Alaamer/medium-articles-posts-with-content

0

null

0

null

2018-08-26

2018-08-26 22:47:03

2018-08-26

2018-08-26 23:12:56

5

true

en

2018-08-26

2018-08-26 23:13:07

9

120550ea20a7

3.403145

7

2

0

While ghost hunting is our favorite form of paranormal investigation, we also enjoy a good alien conspiracy. We recently watched an episode…

5

Alien Attraction Credit: Pixabay While ghost hunting is our favorite form of paranormal investigation, we also enjoy a good alien conspiracy. We recently watched an episode of Ghost Adventures which felt like a crossover with Ancient Aliens, as the Ghost Adventure crew investigated a UFO sighting hotspot. In the episode, titled Stardust Ranch, one of the UFO witnesses said that he believed one reason for the frequency of UFO sightings was the proximity to a nuclear power plant. UFOs, the witness posited, must either collect radioactive energy from nuclear plants, or have a mission to monitor humankind’s use of nuclear power. That got us thinking: wait, what? Is that true? Maybe UFO sighting do happen more frequently around nuclear power plants? It’s worth noting that this same witness claimed to have killed eighteen aliens with his katana, so he’s definitely an expert. Anyway, we decided to answer the question: Do people actually report UFO sightings more frequently around nuclear power plants? To investigate this, the first thing we needed was a database of UFO sightings. Luckily the National UFO Reporting Center (NUFORC) maintains an awesome database, including dates, locations, and detailed descriptions of UFOs. There are about ~100k sightings available, but we had trouble downloading them all in bulk. Instead, we focused on 23,665 sightings where the UFOs were classified as “lights”. Unfortunately, the sighting locations are listed as a city and state, so we needed to convert this information into a more quantitative location. You’d think we could just use the cities and an API to look up latitudes and longitudes…but, apparently, most free services online can’t handle 20k requests. Instead, we used zipcode, a Python module which lists zip codes for every US town. Finally, we used this table available on GitHub to convert our zip codes to latitudes and longitudes. Blue: UFO Sightings in the continental US. Image generated with Python using BaseMap We also needed a database of nuclear reactor locations, which we got from the US Energy Information Administration. Again, we converted zip codes to latitudes and longitudes. This data contained the locations of 22 currently operational nuclear reactors. Blue: UFO Sightings in the continental US; Red: Nuclear power plant locations. Image generated with Python using BaseMap Hey! These do look kinda correlated by eye. However, you can also see that UFO sightings seem to trace major cities, so we suspected that both nuclear reactors and UFO sightings just follow population density. To explore this possibly confounding variable, we found a database of US population density from data.world in the form of census population data by zip code. Blue: UFO Sightings in the continental US; Red: Nuclear power plant locations; White: Population density. Image generated with Python using BaseMap Yeah, they kinda all look correlated. Let’s do some ~~statistics~~ to make sure. To take a closer look at this we wanted to answer the question: are there more UFO sightings per person in the area around a nuclear reactor than around a random location with a similar population density? We randomly selected a large set of locations and used rejection sampling to recreate the same distribution of population density as seen with the nuclear reactor sites. The metric we used to measure population density was summing the population of people within a 10 km radius of a given location. After we had our set of locations, we normalized the population densities with the number of UFO sightings within the same 10 km radius. We then average the UFO sightings per person for all of the nuclear reactor sites and compared that to the UFO sightings per person for all of the control sites. We found: Average sighting concentration within 10 km of nuclear reactors: 5± 6 sightings per 1 million people. Average sighting concentration within 10 km of control sites: 5± 6 sightings per 1 million people. In other words, they’re literally the same. There is no (interesting) correlation between nuclear power plant sites and UFO sightings. However, it’s true that nuclear power plants and UFO sightings appear correlated at first glance, if you do not account for population density. This is one example of why it’s so important to account for confounding variables in statistical analysis. All this said, the only alien sighting involving a katana (that we know of), did occur near a power plant. Perhaps one day we’ll have more data to better substantiate this tantalizing correlation.

Alien Attraction

110

alien-attraction-120550ea20a7

2018-08-27

2018-08-27 02:19:09

https://medium.com/s/story/alien-attraction-120550ea20a7

false

681

null

Python

python

Python

20,142

Ashley Villar & Alex McCarthy

We apparently only write about ghosts and aliens…statistically.

e7cd572a7888

astrovav

4

3

20,181,104

null

0

null

0

null

2017-02-20

2017-02-20 20:02:05

2017-09-08

2017-09-08 15:41:26

7

false

fr

2017-09-11

2017-09-11 07:56:54

29

1207659f6e3b

13.717925

0

Avez-vous vraiment envie, en 2017, de faire confiance à votre agence de com data-driven ?

5

Le marketing de l’influence, l’utopie de trop ? Avez-vous vraiment envie, en 2017, de faire confiance à votre agence de com data-driven ? Et un message vocal de plus ! “Il faut qu’on se parle, on a lancé une nouvelle practice !”. Encore un sale coup du dircli de votre bonne vieille agence de com. Après avoir rempli le pèle-mêle de votre boîte email de quatre relances pour vous inviter à lire son dernier whitepaper sur le data marketing affinitaire, l’autoproclamé pape du branding joue le all-in et vous envoie sans détour une proposition commerciale pour venir crafter votre 2020 strategic masterplan. No kidding, of course, car l’agence de com’ joue cette fois la carte data-driven, posée par le DGA en charge du digital subitement propulsé à la tête de la nouvelle cellule UX/IA : “vos prochaines campagnes seront pushées par un réseau de micro-influenceurs, avec des messages tirés d’insights marché, et donc dans une démarche totalement roi-isée”. Comprenez que c’est prouvé : pour 1€ investi, on se met 1€20 dans la poche !”. Pourquoi se risquer à faire de l’influence marketing “data-driven”, alors que tous les derniers projets de com digitale n’ont rien apporté à la marque ? Et si c’était le devis de trop ? Il y a 3 ans, on vous a fait le coup du guest blogging, puis de la communication inclusive, et maintenant du brand IQ … ! Résultat de la dernière campagne, à peine +0,6% sur les ventes pour un onéreux re-déploiement de votre mix media dans une logique multicanale. Alors, data-driven ou pas, cette fois, vous ne serez pas dupe, et vous préférez “revenir aux fondamentaux” pour continuer à travailler les grandes thématiques de votre marque corporate avec du brand content. Et même quelques vidéos en prime pour redorer l’image de votre marque employeur sur YouTube, et aider les RH à “sourcer les meilleurs talents”. Au moins, cette décision passera tranquillement au prochain codir ! Et pour la brand influence data-driven, on verra avec le futur dircom qui vous succédera quand vous aurez enfin récupérer le marketing dans votre escarcelle — une fois le 2020 strategic masterplan mis en oeuvre. Du coup, merci beaucoup à l’agence, ce n’est pas notre priorité, on se rappelle l’année prochaine. Don’t believe the hype? L’influence marketing, une vaste supercherie technologique revenue à la mode grâce aux agences d’innovation data-driven ? Vous, au moins, vous n’êtes pas tombé dans le panneau ! Si les articles du site corporate n’ont pas généré de trafic l’année dernière, c’est parce que l’agence media a mal fait son travail de promotion de vos contenus sur les blogs spécialisés de votre secteur. A quoi bon lancer une nouvelle initiative pour identifier de nouveaux canaux et de nouveaux modes de communication pour votre marque ? De toute façon, votre problématique est avant tout B2B2C, rien à voir avec les pseudo-succès de petites marques présentés tous les mois dans Stratégies. TRVSN, une marque street-wear premium qui doit son succès à la proximité de sa communauté Pourtant, pendant que votre café se prépare, vous tombez sur le dernier post Linkedin de l’agence qui avait voulu obtenir le budget de votre marque l’année dernière. “Grâce à Instagram et à notre réseau d’influenceurs, nous avons généré plus de 400k leads pour une célèbre marque française de cosmétique — qui a alors gagné 3% de parts de marché”. Pas mal ! Mais bon, c’est du B2C grand public, forcément ça plaît aux jeunes qui passent leur temps sur leur smartphone, et ça crée de l’engagement … Pas pour vous ! Deux minutes plus tard, vous tombez sur une success story d’influence marketing B2B. Finalement, le doute s’installe. Vous aviez déjà mis cinq ans à vous mettre au brand content. Maintenant, vous avez peur de rater le train de l’influence marketing ! Après quelques heures de recherches sur Google pour faire le point, vous faites volte-face face à la déferlante des articles sur le micro-influence marketing, et vous rappelez l’agence ! Et vous voilà lancé dans un programme d’influence marketing sur un an. Une fois le devis signé et le bon de commande validé, une question vous vient à l’esprit : “mais… comment je vais démontrer l’impact de ces publications social media sur mes ventes ?”. Ca y est, vous êtes entré malgré vous dans la spirale infernale de la crise de l’influence. Malgré vous, vous êtes rentré dans la spirale infernale de la crise de l’influence. La spirale est infernale, car vous avez mis la charrue avant les boeufs. Avant de faire de la com’ via des influenceurs, quels qu’ils soient, il faut déjà savoir quel message on veut véhiculer. Et même avant cela, il faut savoir si le produit / service qu’on veut développer correspond vraiment aux attentes d’une communauté donnée. Attention à bien comprendre ce qu’on veut faire de sa marque, avant de déclarer que l’on fait de l’influence. L’influence de marque, ce n’est pas un nouvel élément du marketing-mix, c’est une posture organisationnelle qui recentre l’action de la marque sur les attentes des consommateurs. La marque influente parvient à adapter sa proposition de valeur à la perception de la valeur chez ses prospects, et non à convaincre d’acheter quoique ce soit par le biais de gens reconnus dans une communauté — auquel cas la marque s’expose à un epic fail potentiel de vente ou de réputation. La stratégie de communication centrée sur des micro-influenceurs doit ainsi une conséquence d’un bon positionnement, mais jamais une cause de l’influence de la marque. La crise de l’influence : un problème de définition Pourquoi ? Parce que votre agence s’est trompée de définition. Ce qu’elle appelle l’influence, c’est le tracking d’un potentiel de scoring social — qui ne reflète pas la capacité à faire changer le comportement ou l’avis de quelqu’un. Ce qu’on appelle l’influence, c’est la capacité à faire changer le comportement de quelqu’un — et non un potentiel de scoring social, semblable à celui d’un simple panneau publicitaire Les influenceurs propulsés par les réseaux sociaux sont généralement de simples espaces media complémentaires aux canaux traditionnels des marques (TV, radio, display, print…), qui se retrouvent eux-aussi souvent à la loi du marché et ainsi parfois achetés à prix d’or. Les KPIs associés à l’action de ces vitrines ambulantes sont d’ailleurs exactement les mêmes que ceux gérés historiquement par les régies : reach, relevance & resonance. Ces “3Rs” sont peut-être aujourd’hui mieux reconnus via les metrics de social engagement (likes, shares…) et par la toute-puissance du Klout score. Rassurez-vous, l’influence de la marque n’est pas morte-née avec l’émergence des influenceurs ! Paradoxalement, la marque n’a jamais autant été confrontée à la problématique de l’influence afin de proposer des solutions adaptées aux attentes de segments toujours plus atomisés et aux offres de concurrents toujours plus verticaux. Le succès rapide de marques comme BlaBlaCar, Algolia ou Sonos démontre davantage une capacité à s’adapter aux besoins fluctuant de communautés en constante évolution, qu’une capacité à mobiliser un réseau d’influenceurs — aussi microscopiques soient-ils. Le paradigme de l’influence doit aujourd’hui être renversé pour laisser la place à l’influence des consommateurs. Plutôt que de se construire un réseau d’influence(urs), la marque doit accepter de s’intégrer au mieux dans des communautés en constante évolution dans notre société liquide. La marque doit désormais faire face à la crise de l’influence en acceptant de se laisser influencer par ses propres communautés de consommateurs. C’est en lâchant prise qu’elle peut s’ouvrir à ses propres clients — via une démarche maintenant traditionnelle de feedback management, mais surtout à ses prospects à géométrie variable — dans une démarche d’écoute des conversations qui gravitent autour de la marque. En moyenne, chaque internaute génère plus de 300Mo de données par jour. Une grande partie d’entre elles sont conversationnelles, et accessibles aux marques La crise de l’influence, amplifiée par la modernité liquide Nous vivons désormais dans une société liquide, matérialisée par une reconnaissance toujours plus forte des individus par leurs actes de consommation. Zygmunt Bauman décrit cette modernité liquide par le prisme du réseau : “La tendance à substituer la notion de réseau à celle de structure dans les descriptions des interactions humaines contemporaines traduit parfaitement ce nouvel air du temps. Contrairement aux « structures » de naguère, dont la raison d’être était d’attacher par des nœuds difficiles à dénouer, les réseaux servent aujourd’hui autant à déconnecter qu’à connecter”. La modernité liquidité impacte directement l’action des marques via le phénomène bien connu de la “crise de l’attention”. L’ère du Web social dit 2.0 a introduit un paradigme d’ubiquité : les nouvelles technologies de ciblage, la multiplicité des canaux et la facilité d’accès aux analyses de données utilisateurs à grande échelle (big data), toutes ces révolutions techniques permettent aux marques de parler à tout moment à leurs consommateurs, au bon moment, avec le bon message. Pourtant, malgré cet impressionnant arsenal technologique, les marques n’ont jamais été autant confrontées à l’inefficacité de leur communication. Les success stories de marques devenues iconiques comme Apple, Redbull ou Hush Puppies ont souvent été décrites comme conséquences uniques d’une mobilisation d’un establishment ou d’un groupe d’individus alphas (sportifs, mannequins, artistes…) capables de mobiliser subtilement l’attention des foules autour d’un produit donné. Proches du sponsoring mais différentes par leur ingéniosité, ces opérations de branding avaient été savamment préparées pour faire converger la valeur du produit avec l’idéal social renvoyé par le groupe d’influenceurs sélectionnés. Efficace dans des années 90s encore peu marquées par le phénomène Internet, cette stratégie a perdu de son crédit avec la fin du règne des Influentials tant appréciés par le Time Magazine. La loi du petit nombre, théorisée par Malcolm Gladwell dans son ouvrage Tipping Points autour de supermen & wonderwomen à l’aura suffisamment puissante pour faire basculer l’opinion du grand public, est remise en cause par la nature-même de la postmodernité. L’éclatement des sources d’information induit par Internet — tout autant caractérisé par la multiplication des fake news que des plateformes de comparaison— laisse aujourd’hui peu de place à la formation d’un référentiel commun de consommation, et donc peu d’espace aux leaders d’opinion décrits par Gladwell. Au contraire, l’influence est bien plus induite par des réseaux resserrés, les close networks que présente Duncan Watts dans Six Degrees, que par des meta-influenceurs. Le développement des réseaux sociaux conjugué à l’accroissement de l’image de soi dans notre société liquide a démultiplié l’impact des échanges avec nos cercles affinitaires sur notre mode de consommation. Selon Duncan, un individu connaissant quelqu’un possédant un iPad aura 14x plus de chance d’acheter lui-même qu’un autre individu ne subissant par la pression sociale de son réseau formé “d’everyday people” — et ce quelque soit l’exposition médiatique à laquelle il est confronté. Is the Tipping Point Toast? Don't get Duncan Watts started on the Hush Puppies. "Oh, God," he groans when the subject comes up. "Not them." The…www.fastcompany.com Finalement, seul notre réseau de proximité vient réellement influencer notre comportement, et nous pousser à l’acte de décision ou d’achat. De fait, nous sommes tous à la fois influenceurs et influencés. Les Influentials, qui travaillent avec les marques, sont uniquement des accélérateurs de décision, à l’instar des autres dispositifs autour de la notoriété (display, print…). Le challenge de la marque en quête d’influence est donc d’être bien perçue par un ensemble d’individus qui reconnaissent totalement la valeur d’usage du produit / service, au point d’amplifier cette valeur perçue auprès de leur propre communauté de proximité. L’influence de marque se définit alors par la puissance de réalisation du moment zéro de vérité d’un ensemble d’early adopters donné L’influence de marque se définit alors par la puissance de réalisation du moment zéro de vérité d’un ensemble d’early adopters donné, c’est-à-dire par la capacité d’individus à reconnaître (avant l’achat) la valeur d’une réponse commerciale apportée face à leur besoin de différenciation. Ce besoin peut être de nature technique (désir de meilleurs performances), sociale (désir d’appartenance à un groupe inédit) ou inspirationnel (désir de devenir autre). In fine, le choix rapide fait par les early adopters vient alors nourrir les envies de leurs propres réseaux de proximité, entraînant une influence naturelle de la marque en faisant changer le comportement d’un ensemble de consommateurs finaux. Renverser le système de l’influence Pour se définir comme influente, la marque doit être en capacité de faciliter la reconnaissance de sa proposition de valeur par une communauté d’early adopters. L’enjeu n’est pas tant la rapidité de cette reconnaissance que l’adéquation totale entre la valeur proposée et la valeur perçue par ces early adopters. Blablacar, champion du covoiturage, a bien mis presque 10 ans pour devenir une licorne ! Nokia avait bien réussi à vendre plus d’un million de consoles N-Gage, ce qui n’a pas empêché l’échec commercial de son produit — à cause d’un différentiel trop important entre les attentes (un appareil de poche multi-usage) et la proposition de valeur (un environnement fermé, à l’expérience non-personnalisable comme l’iPhone et ses applications tierses). Le “piège du device” : beaucoup de fonctions, trop peu peu de simplicité Avec la N-Gage, Nokia est tombé dans le “piège du device” avec trop de fonctions embarquées inutiles au grand nombre, plutôt que de concevoir un produit “magnétique”, capable d’attirer la masse par une expérience-utilisateur (UX) adaptable aux besoins de chacun, sans remettre en cause les fondamentaux du produit. Le développement du produit n’a pas pris en compte les besoins réels du marché, entraînant un décalage entre la proposition de valeur (un device ultra-complet) et la valeur perçue du marché (un device de geek, trop complexe), et au final un échec de l’influence de la marque. Pour devenir influente, la marque doit être capable de lâcher prise et de se laisser influencer par ses propres consommateurs. L’influence n’est pas canalisable par des meta-influenceurs, quelque soit la nature du produit ou du service développé. Pour devenir influente, la marque doit être capable de lâcher prise, et de se laisser influencer par les consommateurs, c’est-à-dire par les individus qui formeront son futur marché. Lâcher prise, c’est accepter de renoncer à ses propres intuitions pour avancer dans le processus d’innovation — en se reconnectant au maximum aux signaux envoyés par les futurs clients de la marque. La principale difficulté réside dans l’inaudibilité apparente des signaux envoyés par les consommateurs aux marques. Ces signaux sont par nature non-structurés, puisque constitués de langage naturel, et sont surtout éparpillés sur des canaux très différenciés complexes à regrouper dans une seule et même base de données : Publications spontanées sur les réseaux sociaux : Instagram, Twitter, Facebook, Snapchat, WeChat … Adressées aux cercles particuliers des individus, ces publications mettent régulièrement en scène les points d’enchantement et de frustration associés à l’expérience de produits ou de services Publications envoyées directement aux marques sur les réseaux sociaux, dans une logique de découverte ou de SAV (typiquement via Twitter, Messenger ou Whatsapp) Reviews client sur les sites de e-commerce ou de comparaison (Amazon, Tripadvisor, Beauté-Test…) Conversations directes avec les marques enregistrées sur les CRM (historiques d’emails, de chatbots, transcripts de call centers…) Résultats d’enquêtes / surveys envoyés à de larges bases de prospects ou réalisées in situ (via des réponses à des questions ouvertes) La plupart de ces données sont par nature privées, car propriétaires des marques. Elles constituent en ce sens un référentiel privilégié d’étude, qui regorgent d’insights sur le comportement, le lifestyle et les attentes des consommateurs — et parfois de micro-segments spécifiques. Afin de donner du sens à cette masse de données conversationnelles, la marque doit désormais être capable de s’équiper de technologies capables d’aider l’humain à compiler tous les signaux faibles. La prise de parole des consommateurs est le plus souvent elliptique et implicite, surtout lorsqu’elle est spontanée et non-adressée directement à la marque. L’innovateur humain n’a plus aujourd’hui la capacité cognitive de processer seul l’ensemble des données auxquelles il est confronté — par leur complexité et par leur multiplicité. Heureusement, l’intelligence artificielle vient répondre au challenge de la manipulation de larges sources de données non-structurées. Les technologies d’I.A. sémantiques viennent aider l’humain à identifier les signaux faibles en allant identifier des comportements saillants dans des sous-ensembles que l’homme n’aurait pas pu facilement imaginer. Quasiment infinies, ces croisements entre sujets mentionnés et sous-ensembles spécifiques sont générés par des algorithmes capables de clusteriser des éléments de langage, et donc de révéler des comportements dans des communautés spécifiques. Le clustering sémantique généré par l’I.A. permet de déceler immédiatement les thématiques mentionnées dans un ensemble hétérogène de conversations autour d’un produit, d’une marque ou d’un concept théorique Par exemple, en remarquant que les discussions sur les crèmes contre l’acné sont associées aux risques du cancer sur des forums de discussion entre jeunes mamans, alors que les réseaux sociaux plus contemporains mettent davantage en avant des problématiques de rougeur ou d’irritation de la peau. De même, il est possible d’isoler des comportements spécifiques aux canaux, lorsqu’une marque est mentionnée en ligne, en identifiant les principaux sujets associés à la marque dans les articles. Un exemple de clustering sémantique (les thématiques en ordonnée) croisé avec une variable d’analyse : le canal d’origine des conversations (Facebook, Twitter…) Ces analyses portées par la technologie demeurent néanmoins stériles si elles sont réalisées sans but déterminé, étant donné l’infini des possibles en terme de réalisations d’étude. La principale force de l’approche sémantique intégrée et augmentée par l’intelligence artificielle vient de sa capacité à venir conforter, ou infirmer, l’intuition de l’innovateur. La puissance de calcul exponentielle de nos machines permet aujourd’hui à n’importe quel brainworker (dirigeant, consultant, créatif …) de lancer presqu’en temps-réel des analyses de perception d’un produit, d’un service, d’un concept ou même d’un événement. Pour entrer dans la course à l’influence, la marque doit désormais accepter de vivre dans des tourbillons d’influence de ses consommateurs Finalement, pour entrer dans la course à l’influence, la marque doit désormais accepter de vivre dans des tourbillons d'influence de ses consommateurs, et ainsi se mettre en mesure de capter les conversations de ces derniers. L’essor des réseaux sociaux et des plateformes de reviews fut la première pierre angulaire de la multiplication des prises de parole de consommateurs. Désormais, le développement de nouvelles technologies de transcription vocale (natural-language understanding) constitue la formation d’un gigantesque flux supplémentaire d’informations que la marque peut venir analyser dans un unique référentiel. C’est uniquement grâce à ce saut en avant technologique que la marque est désormais en mesure de décrypter des insights sur des communautés de plus en plus fine. Tout l’enjeu pour la marque, notamment la marque naissante, est maintenant de participer à générer ces tourbillons d'influence, en provoquant des partages d'expérience. La génération de conversations ne requiert cependant par une notoriété au préalable, ou un budget marketing démesuré. Bien au contraire, elle peut se faire grâce à des actions de marketing micro-ciblées comme de micro-événements aux retombées social media importantes (tels que les happenings) qui sont autant de matière à venir analyser pour bien comprendre le lifestyle et les attentes de consommateurs qui vont devenir les principaux prescripteurs d’innovation pour la marque. Appendix : les quatre règles de l’influence liquide Dans un contexte de modernité liquide, caractérisée par un environnement socio-économique en constante évolution, la marque doit aujourd’hui renverser le système de l’influence. En se laissant elle-même influencée par les conversations des consommateurs eux-mêmes, la marque est alors en mesure d’innover réellement dans une approche customer-centric, de maximiser la valeur perçue de sa proposition de valeur, et d’entrainer la masse autour d’une première communauté d’early adopters. Dans ce contexte de modernité liquide, caractérisée par un environnement socio-économique en constante évolution, la marque doit aujourd’hui renverser le système de l’influence Exigeant, cette vision de l’innovation implique naturellement l’utilisation des dernières technologies d’intelligence artificielle de gestion des données conversationnelles, mais une remise en cause organisationnelle et managériale, pour rendre la structure capable d’être suffisamment agile pour remettre perpétuellement en question les convictions de ses propres collaborateurs. #nopainnogain La marque influente de demain sera néanmoins la marque capable de respecter scrupuleusement les quatre grands préceptes de l’influence liquide. Capter les conversations de consommateurs en identifiant toutes les sources de prise de parole existantes et accessibles à la marque — en complétant les avis d’experts métier par des dispositifs de web listening, de speech-to-text ou d’ethno-marketing Structurer les conversations pour leur donner du sens — en créant un référentiel d’étude unique (via une DMP) et en exploitant de nouvelles technologies d’analyse intelligentes (comme l’I.A. sémantique) Amplifier la génération de conversations, en suscitant les conversations d’individus grâce à des actions de marketing affinitaire capable de générer de l’engagement et donc des panels d’étude élargis Recommencer sans cesse !

Le marketing de l’influence, l’utopie de trop ?

0

avez-vous-vraiment-envie-de-faire-confiance-à-votre-agence-de-com-1207659f6e3b

2018-03-23

2018-03-23 06:37:56

https://medium.com/s/story/avez-vous-vraiment-envie-de-faire-confiance-à-votre-agence-de-com-1207659f6e3b

false

3,357

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Matthieu Danielou

AI Product manager. An overview of business capabilities offered by artificial intelligence algorithms.

83e41e0e0135

mdanielou

92

144

20,181,104

null

0

null

0

null

2017-10-23

2017-10-23 02:51:03

2017-10-23

2017-10-23 07:56:22

0

false

en

2017-10-30

2017-10-30 02:27:41

2

12078d1c1f5

3.271698

0

Let us look around what kind of technology we are surrounded by today. How many electronic devices are you using right now? I own a…

5

Does Modern Technology Teach People Creative Skills or Artificial Intelligence to Survive in the Modern Society? Let us look around what kind of technology we are surrounded by today. How many electronic devices are you using right now? I own a smartphone and a laptop. Do you own a car, a television, a computer, an electrical watch, a digital camera, a tablet or a smartphone? Majority teenagers, young adults, middle-aged adults and elderly own at least two of the items that I just mentioned because we live in a modern society that produces technology to become more affordable and profitable than before. Even young children have access to electronics as distractions; thus, parents could have time to watch their shows or do work. I am not trying to say anything about parents because I witness this situation myself at home. In order to write this post, I am required to use a computer with Internet access, which I would need experiences to function a computer. When I use my smartphone, I would search up anything that I do not understand; for example, I would search how to spell a word and its definition instantly, instead of flipping through the pages of a dictionary book. To own a electronic device, it may have made me lazy and wanting to save time because I could of look for the word in the dictionary by taking my time. If there is a question that you may be curious about, the Internet could answer it, but it is not highly recommended if it is related to education. Technology could calculate your work for you, but it does not mean you would understand what the functions were plugged in the answer. Unless, someone who is a professional or highly educated explains the steps in the comments. Another way to see technology is some people would utilize electronics for videography, cinematography, photography and the impact of technology, which influences arts and culture. The digital art culture is also growing with the words spreading through social media. Technology could influence people in many ways from knowledge to fake news, depending how they utilize them as dependent assistants or building tools; thus, I want to challenge this idea by arguing for both sides in future posts. Introducing my debate question, I will be informing and breaking down pros and cons for this question to all my readers: “Does modern technology teach people creative skills or artificial intelligence to survive in the modern society?” This is a reasonable question that people would be arguing about because they may be smarter or dumber, and stronger or weaker than what advanced technology thinks and operates. Human beings may create machines to serve people and become humanized because researchers think Artificial Intelligence (AI) could help the world, according to an article from Business Insider called “18 artificial intelligence researchers reveal the profound changes coming to our lives.” For example, Pieter Abbeel, a computer scientist at the University of California, Berkeley, says robots will keep us safer, especially from disasters; for example, the nuclear meltdown at the Fukushima power plant. Another example is Michael Littman, a computer scientist at Brown University, says we need to rethink how we value people because AI exists to work for what human beings did as labor, and human beings reflect how labor benefits AI the take over their work. Computer researchers agree to blessings of modern technology and the decision for making AI’s existence. Human beings created technology in the first place. AI is surviving because technology exists, who are created by human beings. Another instance, modern technology could teach people creative skills because they would be able to reflect their creativity, innovated skills, team sportsmanship and culture that modern technology cannot teach them, according to Jack Ma, one of the richest and successful entrepreneur in Asia, from the South China Morning Post. People have stronger souls and feelings, and learn from their values and beliefs that technology does not have. Modern technology could teach us to think about our future and find another life in another planet. Human beings explore themselves for they like and favor with the help of technology because they would research on what attracts to acknowledge and adventure outside with a G.P.S. device. Some may disagree on how modern technology should not be a skill because it is not part of humanity and allow it be smarter than human beings. We should train human beings to be faster and better than computers with their open-minded knowledge and critical thinking skills because this is how we work our human brain. We mentally stress and physically express our feelings everyday. Can technology react like human beings? No, but yes if human beings make things happen. Overall, this debate question conflicts how human beings would rely on their intelligence and personal skills. People have different skills and talents from verbal to action that we appreciate to have, but we should not allow modern technology conquer or decide for what human beings represent as race, ethnicity, gender and religion. Sadly, we may need technology to survive in today’s society as an excuse for communication and the purpose of mathematics and science.

Does Modern Technology Teach People Creative Skills or Artificial Intelligence to Survive in the…

0

does-modern-technology-teach-people-creative-skills-or-artificial-intelligence-to-survive-in-the-12078d1c1f5

2017-10-30

2017-10-30 02:27:42

https://medium.com/s/story/does-modern-technology-teach-people-creative-skills-or-artificial-intelligence-to-survive-in-the-12078d1c1f5

false

867

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Daphney

null

12e55b397900

xyu4

0

1

20,181,104

null

0

null

0

null

2017-09-05

2017-09-05 19:18:35

2017-09-05

2017-09-05 19:46:28

1

false

en

2017-09-16

2017-09-16 00:12:55

2

1208a049b9f1

1.645283

9

1

0

Human Skeleton Keypoints Dataset consists of 300,000 images with over 700,000 people.

3

AI CHALLENGER GLOBAL CONTEST IS ON! I’m so excited to announce that our #AIChallenger global competition platform is now officially online and all datasets are ready for download at challenger.ai. Large open Datasets include: Human Skeleton Keypoints Dataset consists of 300,000 images with over 700,000 people. Chinese Language Image Caption Dataset consists of 300,000 images with 1.5 millions Chinese captions. The English-Chinese Machine Translation Dataset consists of over 10 millions Chinese and English paired sentences. US$300K Prize for 5 competitions include: Human Skeletal System Keypoints Detection Scene Classification Image Captioning (Chinese) English-Chinese Simultaneous Interpretation English-Chinese Machine Translation. Free GPU Resources Free GPU resources to those who have limited budget, but are passionate about AI and ready to take actions. Connect with AI talents around the world. Since we announced AI Challenger on August 14, talents and institutions have registered include: Students from Cornell University, Georgia Tech, NYU, University of Cambridge, Imperial College London, Karlsruhe Institute of Technology, Ecole Nationale des Ponts et Chaussees, University of Wollongong, Waseda University, Tsinghua University, Peking University, Chinese Academy of Sciences, Shanghai Jiao Tong University, Fu Dan University, Hong Kong U. of Science & Technology, Chinese U. of Hong Kong, National Taiwan University and more! Developers from Microsoft, GE, Intel, eBay, Micron, BNP Paribas, Baidu, Alibaba, Xiaomi, Sohu, Qihu360, ZhongAn Insurance, China Mobile, China Telecom, DeepGlint, UISEE, Mobike and more Champions and leaders from other top AI competitions such as Kaggle and Tianchi’s. Winners from AI Challenger will have chances to job/internship or obtain investment opportunities with the organizers, present at top conferences, and receive advice from top experts such as former Microsoft Research senior researcher Yi MA, Former Google Senior Staff Research Scientist Dekan LIN, and Megvii Technololy Chief Scientist Jian Sun. Competition Timeline Training dataset is available for download after 10:00AM, September 4, 2017. Open for registration until 23:59:59 on October 31, 2017 Final ranking will be determined by the scores at 23:59:59 on December 3, 2017. Top 5 teams on the final leaderboard of each major competition will be invited to present their work onsite at AI Challenger finalist award event in Beijing in midÂ December. Each competition will also have bi-weekly or weekly prizes. Please help to pass the word, and more importantly, join AI Challenger, and show your talent to the world! All at challenger.ai

Large open Datasets include:

159

large-open-datasets-include-1208a049b9f1

2018-04-03

2018-04-03 23:04:49

https://medium.com/s/story/large-open-datasets-include-1208a049b9f1

false

383

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Kai-Fu Lee

AI Expert, CEO of Sinovation Ventures (创新工场), former President of Google China, Author of “AI Superpowers”

13ba09f08ed3

kaifulee

21,852

52

20,181,104

null

0

null

0

null

2018-02-16

2018-02-16 20:07:45

2018-02-16

2018-02-16 20:43:40

0

false

en

2018-02-16

2018-02-16 21:02:20

0

120bbc62a1b7

1.279245

0

I can’t stop. Love doing deals.

5

Deal Junkie — can’t get enough… I can’t stop. Love doing deals. Most importantly, I am learning like mad. I have never taken a company global before and there are so many things to learn…so many deals to make. I am still trying to figure out what is our best path and where the big hit is going to come from. Our art technology crosses so many segments. We are in a good position. Numerous quotes and invoices out this week. My processes are getting super tight. Mostly, what I absolutely love, is that my entire job is usually one on one human interactions trying to help each other grow and evolve. For example, I just finished a demo with the GVHA (Great Victoria Harbor Authority) and my new friend there, Tom, is a stand up guy. He is innovative and has vision. Great! We are helping him evolve a landmark building with public architectural art using the world’s most advanced sound to light mapping platform. He needs to draw people down in the winter. We are offering a beautiful solution. It is early days, and there are many layers to go through, but I am positive that Tom will champion it through. We are making a strong, business case with the help of the North American Director of the global MK Illumination who does that exact thing for a living. Every dollar invested in public art comes back. In many cases it comes back big time — 3–4 fold. All of this is in my back yard. I live in Victoria, BC Canada. The Inner Harbor is one of the most romantic places in the world. It has magic and charm. We will be adding to make it even more beautiful, fun, and engaging. We have a grand vision to create a world class light show that brings hundreds of thousands of people down throughout the winter. That is a great deal. That is what I am working on all day long. How can you not love it?

Deal Junkie — can’t get enough…

0

deal-junkie-cant-get-enough-120bbc62a1b7

2018-06-16

2018-06-16 22:08:15

https://medium.com/s/story/deal-junkie-cant-get-enough-120bbc62a1b7

false

339

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Jake West

Entrepreneur. Performing Artist. Teacher (Chief Learner).

2b96cbf519fa

JakeWestCircus

123

15

20,181,104

null

0

null

0

93da5952636b

2018-03-21

2018-03-21 18:22:46

2018-03-21

2018-03-21 18:22:47

0

false

en

2018-03-21

2018-03-21 18:22:47

1

120be3de38

2.015094

1

0

null

5

Predictive maintenance: One of the industrial IoT’s big draws One subset of the internet of things — the industrial IoT — adds new capabilities to operational technology including remote management and operational analytics, but the number-one value-add so far has been predictive maintenance. Combining machine learning and artificial intelligence (AI) with the deep pool of data generated by the flood of newly connected devices offers the opportunity to more deeply understand the way complex systems work and interact with each other. And that can promote predictive maintenance — with the ability to pinoint when components of industrial equipment are likely to fail so they can be replaced or repaired before they do, thereby avoiding more costly damage and downtime. According to Wael Elrifai, senior director of sales engineering and data science at Hitachi Vantara — the company’s IoT arm — one of the complexities of predictive maintenance is that AI-produced models for system behavior have to change over time. He used the example of a Hitachi Vantara railway customer with a 27½-year maintenance contract to illustrate the issue. As train parts age, they respond to stresses differently than they do when they’re new. Because of that, maintenance schedules should be adjusted over time to take into consideration changing failure rates. These schedules can be generated with models that are the output of machine learning, he says. There’s a “bathtub curve” to equipment failure, Elrifai said. At the beginning of its service life, there are frequent failures, but maintenance processes get figured out as time passes, so failures become much rarer. “And then, of course, end-of-life — it starts to fail a lot again,” said Elrifai. This type of AI-produced model can be created for other industries as well, and Hitachi has just released a platform called Lumeda that pulls in IIoT data that data scientists can use to adjust their machine-learning models more precisely. “It’s all about being able to monitor machine-learning-model accuracy after a model goes into production,” said Arik Pelkey, senior director of product marketing. One example is a chemical-manufacturing process. Lumada creates a centralized data pool on which data scientists can experiment, so the process of testing different models against each other means that the company can change its inputs and get a more accurate prediction of what’s going to happen to the chemicals at the other end of the production line. Elrifai and Pelkey said that the biggest impact that evolving machine-learning-model management will have will be on low-margin, high-capital businesses, like heavy industry and transportation. Cars manufactured in the past 15 years generally have a computer on board called OBD-II, which stands for on-board diagnostic, version 2. If you’ve seen a mechanic plug a scanner into a specialized port on your car, they’re probably checking with the OBD-II. A startup called TheCarForce is looking to leverage the data from that computer to help drivers and garages — and ultimately, even manufacturers — alike. CarForce’s hardware is a dongle that plugs into that port and stays there, sending diagnostic data, via a SIM card, back to a central hub. Posted on 7wData.be.

Predictive maintenance: One of the industrial IoT’s big draws

1

predictive-maintenance-one-of-the-industrial-iots-big-draws-120be3de38

2018-03-21

2018-03-21 20:43:06

https://medium.com/s/story/predictive-maintenance-one-of-the-industrial-iots-big-draws-120be3de38

false

534

Insights in the People, Process, Technology and Visualisations of the Data Landscape

null

7wdata

null

The Data Intelligence Connection

yves@7wdata.be

the-data-intelligence-connection

DATA,INNOVATION,AGILITY

7wdata

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Yves Mulkers

BI And Data Architect enjoying Family, Social Influencer , love Music and DJ-ing, founder @7wData, content marketing and influencer marketing in the Data world

1335786e6357

YvesMulkers

17,594

8,294

20,181,104

null

0

null

0

null

2018-08-05

2018-08-05 21:46:16

2018-08-05

2018-08-05 21:48:16

1

false

en

2018-08-05

2018-08-05 21:48:16

0

120bf59111a2

2.490566

1

0

“Make way, make way, ’cause today is the day. New friends would land, oh the games we’d play.” Our star was a bot, but no little dude. If…

5

Mars is ill “Make way, make way, ’cause today is the day. New friends would land, oh the games we’d play.” Our star was a bot, but no little dude. If he stood up straight, he’d be 7'2". Curiosity was his name as it came with no shame. Forth rover on Mars was his claim to fame. And though he did he job and stays curious All the space probing had him a little delirious. Rest assured he was no solo fox, as there were other bots. But all the conversations played the same, which became oh so lame. Our boy Curiosity, needed a change. “Although I call ’em and make that hotline bling The bots roll on and do their thing. They see my call and answer the phone I say ‘b-ball?’ and hear ‘leave me alone’. So I do not wish to phone my own among my own I am alone only distance they have shown it is time to change this tone.” “See birthdays used to be the worst days but this year will not be a replay. This year monotony will not win. This years guests will be human. We’ll smoke some tree and keep it breezy. I cannot wait for Mars third specie.” The red planet was to be explored next Under Elons flag, you know, Space X. They go by many names but they’re all the same Connected they were by histories chain. Explorers, colonizers, the tamers of all that is wild. Horrors, fertilizers, they’d make of Manifest destinies child. At that moment, their space craft began descending. The bot played it cool, though just pretending. “Getting a gift would be oh so tight Even cooler if I got a Lite Brite®” “Beep beep boop beep” said the rover in anticipation. “Finally, birthday guests, from the US nation.” So they did land, 12 in their band. Yet another small step for man. They went on to record, document and even test, worried not of martian inhabitants. For there were none in sight to put up a fight Not one of the 12 had that lite brite® But his cool he could not keep So he spoke with his beeps “Hello friends, one and all, I’m pretty sure, they call you Paul. This is Mars, land of dust and creators The technicalities we’ll get to later My only warning is to keep an eye out for the natives To which I’ve named ‘The bing-bators’ Caution for ‘bing-bators’ I do share As they are neither here nor there, but everywhere The 12 did not listen but went on with the mission. The rover was nice but they could care less, and they did not hear about the indigenous. Making friends wasn’t easy instead quite complex. Here one moment, gone the next. So time went on, and on, and on. 7 days past and no birthday song. The rover was mad it had been a week, and not one game of hide and seek! No games of tag, no slap the bag. This had been a real drag. So the rover roved and went on a search. On a large red cliff he did perch. New friends weren’t fun, he had been fed lies. Then our robo couldnt believe his eyes… Those explorers from the land of burgers and fries Would come to be the colonized. The little martians smaller then a hair, Their poor immune systems could not bare. Now 12 explorers lay so quiet and dead Cut from those that filled natives with lead Seeking real estate on “dem Mars space rocks” But no match for Martian small pox.

Mars is ill

50

mars-is-ill-120bf59111a2

2018-08-05

2018-08-05 21:48:16

https://medium.com/s/story/mars-is-ill-120bf59111a2

false

607

null

Mars

mars

Mars

2,038

Jesus Trevino

I got opinions too!

ce64db1847bc

jesustrevino_29095

19

162

20,181,104

null

0

null

0

null

2018-07-02

2018-07-02 01:10:24

2018-07-02

2018-07-02 01:12:51

0

false

pt

2018-10-01

2018-10-01 05:16:23

0

120c40b82411

5.403774

2

0

Thinking about data science, artificial intelligence, and machine learning, what is the best language? Python or R?

4

Python vs R Thinking about data science, artificial intelligence, and machine learning, what is the best language? Python or R? There are no more codes out there, as there are other articles on this. When I was interested in programming and data science, I was already in physics college, I discovered the world of data analysis and data science is giant and a bit scary at first, and my biggest attention was, what is the better language? R or Python? The answer is really straightforward, you should learn Python and R, also learn about SQL database and the like. But you can have more fun forward, give this chance to Java and Scala for example. But how to learn? You should learn about the fundamentals and control flow structures. This is now geared to the use of data such as query, parsing, filtering, extraction, sorting, aggregation, visualization and exploratory data analysis. The question of the times is not always to learn to learn and to learn the language to care and to learn to know in a faster and more difficult way using the programming language and the fulfillment of such objectives. Programming languages, their characteristics and their paradigms: We know that most languages have many concepts, paradigms and algorithm structures. Programming languages differ by design and different attributes, processing and execution of code (typing and paradigm). Sometimes after being typed the codes are preprocessed before their execution, in other cases the code is executed directly by the compiler (translates the high level to the low level), after it is compiled, executed by a mechanism or by runtime. Interpreted languages are executed on the fly without requiring compilation by an interpreter, for a code prototype they are faster, but may not have some “benefits” of optimization and performance of compiled code execution. But generally compiled languages are strongly typed, that is, all variables and code data are explicitly declared with a specific type and must be consistent during code flow and execution. Compiled languages have variation at compile times to check for errors, and this mechanism also promotes type security. Common paradigms of programming languages: Scripting or script programming: I believe it is the simplest paradigm, where code is written in script files that are executed by the language execution engine, where it is usually the interpreter. When loading and executing a script file through a terminal, for example, it executes the language loop’s REV (read-eval-print loop) interactive environment, where the programmer executes it via the command line. Procedural: organized, coupling code-data scope, is written in order to carry out the code by means of well-defined functions, modules that work together to provide the functionality of the program / application. The procedural name is because these modules have functions that are also called procedures, these functions can be reused in all the code and most of the time they receive an input to generate one or more outputs. With the code organized in modules, each containing its functionality, with interfaces less defined (API’s) it gets. Scope of data visibility to data access to modules and individual functions. Most of the time one will define the data (variables) that are global locations at that scope, this data can be viewed and used by the code anywhere in the program, but the local scope data is understood by and is available in a single function . Object orientation: it is the idea of an object in code, which are created from classes, these objects have independent properties and methods. I find it very difficult to explain object orientation, so I’ll give you an example: let’s create an application to sell shoes, and this will be our object model of the class shoes. The properties of the object are values that characterize it, whereas the methods of the object (functions) are the features or actions. An object shoes shoes class and give you properties like type, brand, size, price and so on. While all shoes in this app will have these properties and each shoe object will have different values. Python Multiparadigma programming language, characterized as a dynamically typed language, with script, procedural, interpreted and object oriented. Python has a very comprehensive standard library. It is multifunctional, it can be used for everything, data science to administration of network systems. web applications, script execution and so on. Python is simple, and it was very easy to learn when you wanted to “swap” C and the community is huge, it has several features and tutorials online. Python is an entry in the field of data science, artificial intelligence and machine learning because of the direct production of powerful and quality packages and modules. These packages can perform exploratory data analysis, statistics, predictive analytics, machine learning, neural networks, deep learning, recommendation systems, and so on. The lack of packages and features has been discussed a long time ago, compared to R. The comparison of R and Python packages usually need to be obtained from the vendor itself or with a package manager like Anaconda, Miniconda or PIP. The lack of a central Python package repository that the R owns is not a bad thing, it may need more involvement when it comes to installing and updating. R It is also a multiparadigma language, it is dynamically typed, scripted, procedural and interpreted, it can also support a type of object-oriented programming. It is considered statistical software, is specialized and suitable for statistics, analysis and visualization of data. Compared to Python, it is far less flexible and diverse and also has a language-focused community. It is not difficult to learn but it is very strange. R differs from Python in its native implementation and supports matrix arithmetic and associated data structures, with vectors and matrices, very close to matlab. Python uses numpy for this. There are people who say that R is superior to Python for both statistics and visualization in terms of packages, implementation details and final aesthetic results for visualization — but that all depends. The CRAN is the repository of many R packages, it is centralized and well “taken care of”, where all the packages available for the R are included and that includes packages for a large number of tasks. But, it’s not so clear almost packages are better for each task, have many good packages, many more than Python, but are highly specialized and obscure. Where already in Python the packages are more direct, easy to identify and begin. IDE’s Both languages have several command line interfaces, which can be through a terminal running the interactive environment. For R the best known is the RStudio, I’m suspected to speak because I find it impressive, programs and tasks can be executed from a simple text file in combination with a terminal running the R environment. Many notebook options are used, and are very popular among R and Python, for the most common R’s are the reporting tools are RMarkdown and KnitR. For Python there is no “official” IDE, I see the most common ones for Spyder and PyCharm (where the communit version is open source), but any editor can be considered an IDE, scripts can also be run through the environment . The most common Python notebook is Jupyter, which I’ve seen people using as an IDE. Which one to use, when and why? In both languages have several packages that are very solid and highly capable, it is not wrong to want to start with R instead of Python. I always say to anyone who wants to start in the area that it is best to start with Python as it is easy to learn and get started quickly. It is important to mention that data science tasks are performed on desktops, notebooks or servers (cloud or distributive systems). Between running a development or production task, where implementations have a unique set of considerations, challenges, and requirements that involve packaging, DevOps, site reliability, monitoring, and so on. And from a requirements perspective, we must also consider the deployments that produce the result in real time (or near) and also if the provision of a task, such as recommendation systems, is prepared for a performance target offline or online. I hope I have somehow helped with small differences between these two languages most commonly used in the data science field. It’s worth remembering that things change over time (not as fast as a framework for JavaScript — laughs), and there will not always be a solution for each case. Experimentation and testing are needed to find an ideal solution.

Python vs R

13

python-vs-r-120c40b82411

2018-10-01

2018-10-01 05:16:23

https://medium.com/s/story/python-vs-r-120c40b82411

false

1,432

null

Python

python

Python

20,142

Natália Raythz

9a58bf9135cd

nraythz

33

4

20,181,104

null

0

null

0

null

2017-12-01

2017-12-01 11:32:44

2017-12-01

2017-12-01 11:34:36

0

false

en

2017-12-01

2017-12-01 11:34:36

2

120c9bb46f4e

0.558491

1

0

For B. Tech, M.tech, BCA, MCA, B.SC, MSC and BE students undergoing an 6 Months/Weeks industrial training program is very important as it…

5

Get Your Dream Job With The Help of 6 Months Industrial Training in Noida with 100% Job Guaranteed For B. Tech, M.tech, BCA, MCA, B.SC, MSC and BE students undergoing an 6 Months/Weeks industrial training program is very important as it is an essential part of their curriculum. The training is basically designed for providing comprehensive knowledge about the special area of interest. The training is beneficial for the candidates in more than one way. 6 months industrial training is provided by Octus technology such as R programming, Data Science, Advance Excel, SAS, ERP, Autocad, 3DS Max, Revit, Big Data Hadoop, Python, Php & Digital Marketing with complete practical experiment. • 100% Job guaranteed • Career Advice • Work on Live Project by Real time Industry Expert Octus Technology B-91, 2ND Floor, Sector-2 Noida -201301, Phone No. : 0120–4570915, 8527111645 Website : www.octustechnology.com Email : info@octustechnology.com

Get Your Dream Job With The Help of 6 Months Industrial Training in Noida with 100% Job Guaranteed

1

get-your-dream-job-with-the-help-of-6-months-industrial-training-in-noida-with-100-job-guaranteed-120c9bb46f4e

2018-04-29

2018-04-29 09:03:58

https://medium.com/s/story/get-your-dream-job-with-the-help-of-6-months-industrial-training-in-noida-with-100-job-guaranteed-120c9bb46f4e

false

148

null

Big Data

big-data

Big Data

24,602

OCTUS TECHNOLOGY

null

f974e8e6c728

rahuloctus

2

1

20,181,104

null

0

null

0

5e5bef33608a

2018-04-19

2018-04-19 18:09:24

2018-04-19

2018-04-19 18:12:24

1

false

en

2018-04-19

2018-04-19 18:12:24

1

120d05d45783

1.916981

1

0

The concepts of big data and machine learning are not necessarily new to businesses world-wide; however, what is rather extraordinary is…

3

The Power of Two: How Big Data and Machine Learning Take Artificial Intelligence to the Next Level The concepts of big data and machine learning are not necessarily new to businesses world-wide; however, what is rather extraordinary is the developing movement in industry to combine the two concepts propelling Artificial Intelligence applications to new heights of learning and adaptability. Big Data in Business Big Data is essentially the realization that all of the information being collected online could be harnessed by businesses. The collection of Big Data was revolutionary because companies became capable of making smarter decisions simply by having more access to data and consequently insights than ever before. Machine Learning in Business Intelligent Automation has changed how many businesses operate as well. Robotic Process Automation is able to take in requests and complete them efficiently. Beyond the standard automation, however, is the ability for a machine to learn beyond its programming. This allows the software to adjust and evolve as new information is brought in, making it more efficient than ever. Through Machine Learning, businesses find that lower-end work is completed faster and processed more accurately than ever before. Big Data + Machine Learning = Faster & Smarter Decisions Alone, both Big Data and Machine Learning have their parts in the future of all business around the world, but together, they become so much more. As a business takes in a variety of collected data, it can be challenging to sort the data into trends that are logical and helpful. Through the use of Intelligent Automation, however, this information can be processed. Through Machine Learning, the data can not only be identified and sorted, but as more information is collected by the software, the machine is capable of learning more from it. This kind of Intelligent Automation learns from the data, resulting in data trends that might have been too complex for most human workers. Businesses that utilize both Big Data and Machine Learning are capable of making faster and smarter business decisions than ever before. The data that comes in and is analyzed by an RPA can provide insights that would have been mere guesses in previous years. WorkFusion, a software company that offers AI-powered products purpose-built for operations professionals offers a Case Study as an excellent real-world example of how these concepts can come together to create tangible business benefits. Knowledge is power, and that statement could not be any truer for those companies trying to succeed in business today. Machine Learning is that power which is required to harness the knowledge gained from Big Data, enabling businesses to make faster and smarter decisions with an efficiency that was not even imaginable a few years ago.

The Power of Two: How Big Data and Machine Learning Take Artificial Intelligence to the Next Level

1

the-power-of-two-how-big-data-and-machine-learning-take-artificial-intelligence-to-the-next-level-120d05d45783

2018-04-23

2018-04-23 17:25:47

https://medium.com/s/story/the-power-of-two-how-big-data-and-machine-learning-take-artificial-intelligence-to-the-next-level-120d05d45783

false

455

Latest News, Info and Tutorials on Artificial Intelligence, Machine Learning, Deep Learning, Big Data and what it means for Humanity.

becominghuman.ai

BecomingHumanAI

null

Becoming Human: Artificial Intelligence Magazine

team@chatbotslife.com

becoming-human

ARTIFICIAL INTELLIGENCE,DEEP LEARNING,MACHINE LEARNING,AI,DATA SCIENCE

BecomingHumanAI

Machine Learning

machine-learning

Machine Learning

51,320

Derek Porter

Technology Enthusiast. Writer, Columnist and Blogger. Known for a good rant now and then. Subscribes to common sense and self responsibility ideology.

d7e06d9152ad

derek.william.porter

51

1

20,181,104

null

0

null

0

null

2018-03-25

2018-03-25 21:59:40

2018-07-29

2018-07-29 05:51:45

0

false

en

2018-07-30

2018-07-30 19:03:17

0

120dad660f32

3.483019

0

When I first joined Blue Apron, data support at the company was in dire straits. Data requests were communicated by sending an email to a…

3

Escape from the “data request” backlog: balancing competing priorities @ Blue Apron When I first joined Blue Apron, data support at the company was in dire straits. Data requests were communicated by sending an email to a group alias, which was copied into Asana. Then, once a week, the team would sit in a cramped conference room to stack-rank new requests in comparison to ongoing or previously requested projects. Immediately after, the team would meet with the entire company — yes, you read that right, the entire company —to discuss prioritization of data requests, and ostensibly, timelines. As you would guess, there were a few drawbacks to this process. First, projects took longer than the team expected because data quality was unpredictable, so our velocity was slow. Second, the backlog of requests grew exponentially as the company scaled. More people = more questions = more emails to “data-requests@.” Third, leaders learned that the best way to get their questions answered quickly was to roll deep to the company-wide prioritization. It was not uncommon to see an entire department walk in to the room, presumably just to make a point. Fourth, the data analysts/modelers/scientists were demoralized because it was impossible to make headway, the list only grew and grew, and they had no agency to direct their own work streams. So what did we do? The first step was to make our prioritization process more opaque — yes, more opaque — by shutting the company-wide meeting down and ingesting requests through a form that included a business justification for their question. The added effort required to defend requests eliminated casual questions, and reducing transparency about our overall pipeline created a state of information asymmetry that improved our team’s ability to negotiate with business leaders. Finally, I had some hard conversations with members of the business about the realities of our team’s resourcing and throughput, which resulted in about 40% of requests being deleted from the backlog. Eventually, the Analytics team migrated away from this reactive data request pipeline to strategic quarterly roadmaps, which I’ll expand upon in another post. But it was amazing to see the change in the team’s momentum and morale after we reduced distractions, established air cover, and provided them with agency to proactively address items coming in through the queue. Although every company varies in its data and analytics infrastructure, resourcing, and talent, it seems that balancing short- and long-term constraints are a constant point of tension for service-oriented functions such as an Analytics team, and at Blue Apron we are constantly working to strike a balance (and, only occasionally, running around like our hair is on fire). Here are some tips that we’ve identified to make things work as we have evolved as a team: Find scalable solutions for repeated problems. For example, our consumer analytics team is extremely short-staffed right now, and dozens of A/B tests for our digital platform are piling up as we analyze initiatives related to our physical product (which have not been A/B tested, and thus require more time/lift). Instead of spending time quickly churning out individual analyses, an analyst on the team is spending the next few weeks building an A/B testing tool that ingests ongoing test groups and then surfaces comparisons between test/control across key metrics such as LTV, churn, order rate, conversion, etc. Because many of our analyses also segment results based on key attributes, this tool will also allow for filtering across those commonly occurring segments to surface interaction effects. The benefit of prioritizing this scalable solution is that it will allow our digital PM’s to self-serve while maintaining an acceptable degree of rigor, while allowing the analyst to dig in to a bigger project that is more engaging that repetition of the same analysis over and over again. Gain cross-functional alignment on the team’s priorities by surfacing trade-offs and being transparent about priorities. When our team was very small (4 analysts), we were faced with extreme resource constraints in terms of what we could work on. One quarter, we were so constrained that we were forced to put together a simple list of 5 projects that we would work on, the estimated headcount for this effort, and timelines for completion. Everything on the list was a big project with (1) a high degree of urgency, where (2) analytical rigor was incredibly important — and, in fact, “analytical support for our S1 filing” was actually not #1 on the list. We then circulated our list to leadership at the company, and asked them to weigh in to the group if they disputed our prioritization so we could have a conversation about trade-offs. Say no — with a smile. In the example above, we had to make some hard trade-offs, and to save us from having the same conversation over and over again, we provided transparent answers to our partners so they could make plans rather than kicking the can down the road. In general, I have found that our business partners appreciate candor around constraints, especially if you offer an explanation about the priorities that are taking precedence so they can escalate to their leadership if they challenge the broader decision. Highlight that the team is a servant to the business, not an individual group. Given that we are a centralized team with embedded partnerships, sometimes it helps to remind folks that our team serves a broader set of needs and that the rising tide lifts all boats.

Escape from the “data request” backlog: balancing competing priorities @ Blue Apron

0

escape-from-the-data-request-backlog-and-balancing-the-competing-demands-on-an-analytics-team-120dad660f32

2018-07-30

2018-07-30 19:03:17

https://medium.com/s/story/escape-from-the-data-request-backlog-and-balancing-the-competing-demands-on-an-analytics-team-120dad660f32

false

923

null

Data Analysis

data-analysis

Data Analysis

4,950

Elizabeth Roodhouse (Roody)

Analytics @ Blue Apron

bb86e922f109

roody

8

20,181,104

null

0

null

0

null

2017-10-10

2017-10-10 20:52:17

2017-10-10

2017-10-10 20:56:24

0

false

en

2017-10-18

2017-10-18 15:27:32

1

120f66ac1248

0.996226

3

0

By Es Lee

5

What is Relationship Science? By Es Lee Relationship science is the study of relationships which combines quantifiable data with scientific tools to observe, analyze, describe, and even predict outcomes of individual relationships. It is a relatively new discipline enabled by modern data gathering and computing technology. In many ways, relationship science is an evolution of traditional social sciences such as psychology, sociology, anthropology, economics, political science and linguistics. Both fundamentally study people and relationships, but relationship science does this at an individual relationship level. In contrast, traditional social science techniques have been limited to studying small groups and/or isolating individual principles, and then using that to infer conclusions about all people. Thanks to the smartphone and devices that allow us to record almost every aspect of our lives, we can now examine the entire lifecycle of almost any relationship. What if you could measure and record every interaction with a past love? Every awkward initiation, every laugh, every kiss, every utterance of “I love you”, every kind gesture, and every attempt to pull away. Now imagine doing that for not only for a single relationship but across all our relationships. By studying our interactions with our parents, friends, lovers, we could gain a better sense of who we are holistically…our personalities, our propensities, our vulnerabilities, and even how we come across to others. This data, aggregated, categorized and studied using modern computing and machine learning techniques, could reveal the inner workings of the human psyche. Ultimately relationship science could advance our understanding of all people and the social sciences in ways never possible before.

What is Relationship Science?

3

what-is-relationship-science-120f66ac1248

2017-10-19

2017-10-19 01:25:25

https://medium.com/s/story/what-is-relationship-science-120f66ac1248

false

264

null

Data Science

data-science

Data Science

33,617

Mei: Messaging Improved

It's All About ME & I. The First Mobile Messaging App with Integrated AI and Blockchain Technology 👉🏼 https://textmei.com/

aed05b83d8fe

meimessaging

12

7

20,181,104

null

0

null

0

8e9bde78121d

2018-09-14

2018-09-14 12:47:50

2018-09-15

2018-09-15 16:13:00

2

false

en

2018-09-21

2018-09-21 10:40:27

2

120faa172a43

3.77956

2

0

Written by Joel Schapero

5

What’s Artificial Intelligence All About? Written by Joel Schapero “Any sufficiently advanced technology is indistinguishable from magic.” — Arthur C. Clarke’s Third Law I really love that quote. How did my bank’s chatbot carry on an intelligible conversation? Was it magic? How did the Roomba learn its way around my living room? How does Amazon know what I want, even before I do? No, it’s certainly not magic — but I think of that quote every time I interact with amazing new technology. I love the feeling I get as I begin to recognize the possibilities and, more so, when drilling-down to discover the previous technologies that were cobbled together to make it possible. A computer can play sounds by reading an MP3 file, or produce, edit, or transmit a JPG image, even though both files are simply numbers. Not so long ago this felt like magic. And what about all the technologies that came together to squeeze yesterday’s mainframe into an iPhone? These “macro” technologies are brilliant evolutions and aggregates of previous technologies, built by giants of the previous generation, who used earlier technologies in brilliant new ways to make… magic. And so it is, I learned, with artificial intelligence (AI), as was explained at John Clements’ Applied Analytics conference. With the help of two Silicon Valley masters, Professors Ikhlaq Sidhu and Paris de l'Etraz, we were shown how these new technologies work. By using new data strategies and applying big data statistical tools — automated aggregates of basic statistics — to our vast collections of data, we can predict behavior, make useful decisions, and create new services. Increasingly, we use those predictions to interact with people knowingly, as if by magic. As I listened, I drifted back to business school, the professor writing the tools of statistics on the chalk blackboard — bell curves, degrees of confidence, least squares, regression, etc. In the 1980s we did all this by hand. Sometimes we used punch cards, inputting data into mainframes, where now-archaic software generated hard-to-explain models. I remember having a great idea back then, I combined semester projects for law and statistics in a single paper. I looked at Supreme Court decisions and built a statistical model that would predict how the court would decide cases — I thought it was a fantastic idea. Unfortunately, both my law and statistics professors did not see the magic. They were not even sure it made sense; they both laughed. The blending of subjects was foggy for them and, as I remember, I got a merciful B. Professor Sidhu showed us how machine learning, as well as analysis using the mathematics of statistics, can be applied to our client data using today’s massive processing power to get a precise understanding of what happened in the past, and then, create tools to predict the future. Massive amounts of data can be fed through an advanced machine learning program called a neural net — to find patterns in seeming chaos — and create rules we can then use to make decisions. It can identify with predictable accuracy if it is a cat or a dog, translate idiomatic language, or recognize a tumor. It can predict the style of shoes you like — and when your old shoes will need replacing. Those predictions allow us to focus our efforts, offer new products and have them in the right place — even before our clients know they need them. Like magic, client data has become a most valuable strategic asset. Professor Sidhu showed us, step by step, how those long-ago hours of hand calculations or mainframes have been replaced with a single line of Python code, running somewhere in the cloud. The data can be obtained in real time, its models applied and predictions made immediately; if done well, and communicated through an interface we can appreciate, it begins to look artificially intelligent. This program allowed us a peek under the hood at what makes big data work, its application, how its use has become a driver of corporate strategies, acquisitions, and a hint of what is to come. Intellectually stimulating, my epiphany during the event was how most of AI is not some rarefied super science, but the systems automation and application of well-understood blocks of learning — applying statistics to data. As for AI, despite Elon Musk’s very public concerns, the Terminator’s Skynet is not in sight yet. But what is, are far better decisions — ones based on accurate interpretation of data, much faster, and applied to everyday decisions. Professor de l’Etraz showed convincingly that smart companies will harness this, create better products and services, and deliver them more efficiently to their clients. They will disrupt their competition — just like magic. Please visit and join the John Clements Talent Community. About the author: Joel Schapero is John Clements’ partner for the Corporate Learning practice in Hong Kong. He was a Director of Insured Financial Structures Ltd., London, a Partner and Head of Corporate Finance for private equity firm Peter Wodtke & Partners (UK) Ltd, also in London, and the Director and Head of Structured Finance at UBS in New York. Joel has a bachelor’s degree from the University of Massachusetts in Legal Studies and Economics, and an MBA with concentrations in Finance and Information Systems. He is a graduate of Harvard Business School’s Advanced Management Program.

What’s Artificial Intelligence All About?

22

whats-artificial-intelligence-all-about-120faa172a43

2018-09-21

2018-09-21 10:40:27

https://medium.com/s/story/whats-artificial-intelligence-all-about-120faa172a43

false

900

Discover Your Full Potential with Looking Glass, a Publication from John Clements

null

johnclementsph

null

John Clements Lookingglass

jcdigitalrenewal@gmail.com

the-looking-glass

LEADERSHIP,CAREERS,MANAGEMENT AND LEADERSHIP,PROFESSIONAL DEVELOPMENT,PERSONAL GROWTH

JohnClementsPH

Machine Learning

machine-learning

Machine Learning

51,320

Shiela Manalo

Writer|Graphic Artist|Video Editor|Musician

8dbb2651e54f

iamsimone02

64

25

20,181,104

null

0

null

0

null

2018-05-11

2018-05-11 13:42:01

2018-05-12

2018-05-12 14:15:38

5

false

pt

2018-05-13

2018-05-13 08:50:29

4

1213c22b8116

5.365409

6

0

Uma abordagem conceitual sobre as nossas interações com as coisas do mundo.

5

Interação humana e a evolução das máquinas. Uma abordagem conceitual sobre as nossas interações com as coisas do mundo. “Estamos descobrindo maneiras melhores de desenvolver software, fazendo-o nós mesmos e ajudando outros a fazerem o mesmo. …” O texto acima faz parte do Manifesto Ágil e é o ponto de partida da nossa jornada sobre a evolução das máquinas que apresento neste artigo, de maneira conceitual e cronológica. Autodesenvolvimento Ano de 2018 — as pessoas tem acesso a ferramentas que mudam a maneira de viver, substituem ou facilitam antigas tarefas. Podemos enxergam através de portais que permitem também escutar, sentir, e falar com qualquer pessoa, e a qualquer hora e em qualquer parte do mundo. A nossa capacidade incrível de autodesenvolvimento nos permite acreditar que, temos potencial para fazer essas alterações e melhorias em nós mesmos ou nas nossas ações para alcançar uma meta desejada. Temos a capacidade única de definir nossos objetivos e metas e decidir os métodos que utilizaremos para alcançar isso. Autoconhecimento Mas essa etapa natural da vida está dando um lugar de destaque para o autoconhecimento, que é a capacidade de investigação interna, que pode nos levar a ser mestres de nós mesmo e, de nos dar acesso privilegiado aos nossos próprios pensamentos. Mas não é tão simples como parece. Se todas as pessoas tivessem realmente uma capacidade de auto conhecimento singular, é muito provável que viveríamos uma experiência de vida em conjunto e melhorada, e toda a raça humana compartilharia talvez, de um único objetivo. Mas essa teoria (não validada) só implica dizer que ainda não nos conhecemos suficientemente e que ainda estamos em uma longa jornada de evolução interna. Auto Interação No momento que você acorda e abre os olhos, a sua visão começa a interagir com o local onde você está, transmitindo mensagens rápidas para o cérebro de que tudo está bem, e que o seu corpo estava apenas em estado de sono. E você começa a interagir com o tudo ao seu redor. A sua pele sente o frio do vento, avisando de uma certa maneira, que você precisa se abrigar. Seus ouvidos estão atentos aos veículos na estrada, e qualquer som de buzina ou de freio, coloca o seu corpo em um estado de alerta anormal, e te conduz a tomar decisões rápidas. O ser humano possui essa capacidade de interagir com o mundo, utilizando as suas habilidades básicas: visão, audição, tato, paladar e olfato. “Viver é estar em constante interação.” Com essa grande necessidade de interação e com os avanços da tecnologia, passamos a viver um mundo de extensões. Nossos celulares por exemplo, são um tipo de backup do nosso cérebro, que nos lembra de atividades importantes. E ao longo do tempo, novas extensões estão sendo criadas e novas maneiras de executar tarefas estão sendo descobertas, com a ajuda desses Gadgets inteligentes. Estamos usando essas máquinas para tornar a vida mais fácil. Estamos descobrindo curas para doenças fatais ou mapeando incêndios florestais com ajuda de drones. Essa necessidade de uso das máquinas, também cria um mercado gigantesco de investimentos, para ver quem tem a máquina mais interativa. Percebemos então, que existe uma necessidade de humanizar as máquinas. Máquinas que proporcionam interações mais humanas, transmitem mais confiança e elevam a experiência da interação. Momento I: O domínio invisível As vezes quando estamos vendo Netflix, ou mesmo escolhendo o nosso próximo livro em uma loja virtual, estamos interagindo com aquele sistema. Em termos, há um controle por parte da máquina, que te atrai ao conteúdo usando todos os recursos humanizados possíveis. O sistema aprende o que você gosta de assistir, falar, ouvir, ler e escutar, e te sugeri como, e com o que você deve interagir enquanto vive. E não podemos falar de interação sem falar de usabilidade, que é basicamente um indicador de qualidade que mede o quão natural e fácil é a interação. “o quanto um produto pode ser usado, por um determinado usuário, para atingir um determinado objetivo, com eficiência, eficácia e satisfação, em determinado contexto de uso”. O especialista Jakob Nielson, também sugere também uma lista de como sistemas devem se comportar para que tenha uma boa usabilidade. Um sistema com uma boa usabilidade, proporciona um relacionamento mais humano. Momento II: Auto desenvolvimento Em um futuro próximo, teremos uma tecnologia tão avançada, que não poderemos diferenciar se estaremos em uma ligação com uma pessoa ou uma máquina. A Google lançou recentemente uma nova atualização da sua assistente virtual, capaz de fazer uma reserva em um restaurante por telefone, e responder a perguntas complexas, baseadas até no conhecimento que a máquina tem sobre você. “ Enquanto alguns consideram que os interlocutores devem ser informados de que estão falando com uma máquina, outros se perguntam sobre a possibilidade de que estes robôs tão persuasivos sejam úteis demais para fins comerciais ou políticos.” Isso pode parecer assustador, mas as máquinas estão adquirindo uma capacidade de autodesenvolvimento que antes só pertencia aos seres humanos. “O objetivo do nosso assistente é ajudar a realizar as tarefas”, disse Pichai. E estas novas funcionalidades serão testadas nos próximos meses, segundo o grupo de Mountain View. Momento III: Auto conhecimento Com os últimos avanços da tecnologia, estamos cada vez mais perto de uma inteligência artificial, capaz de interagir de maneira cada vez mais natural e introspectiva. É como se estivéssemos construindo algo, que tem a capacidade de aprender tudo sobre nós de maneira natural e muito mais rápido. Isso gera também uma discussão ética, que coloca as máquinas em uma posição duvidosa. “Abbass apontou que, ao contrário de Siri, Alexa e outros assistentes de IA, o Duplex da Google fala como um humano convincente. Durante a apresentação, Duplex foi feito para parecer mais humano usando “ums” e “mmm-hmms” durante a conversa, onde era altamente provável que a pessoa do outro lado da linha não soubesse que estava falando com um robô. Ele apontou que isso representa uma preocupação ética e o Duplex deveria se apresentar como um robô”. Momento V: Auto interação O ápice da auto interação entre humano e máquina, se iniciará quando esse relacionamento for natural e fluido. E quando esse momento chegar, as máquinas estarão cada vez mais presente nas nossas vidas e poderão até decidir o que você quer comprar. Na verdade elas saberão de quase tudo sobre você. As máquinas estão caminhando para uma humanização irreversível, e logo, dividiremos o planeta com a próxima raça inteligente e criativa, que como você, conseguirá compartilhar emoções e desejos como qualquer ser humano. Mas essa é uma teoria de inúmeras que rodeiam esse tema. Inteligência artificial (artificial intelligence — A.I., em inglês) é um ramo de pesquisa da Ciência da Computação que se ocupa em desenvolver mecanismos e dispositivos tecnológicos que possam simular o raciocínio humano, ou seja, a inteligência que é característica dos seres humanos. Como você pode ter notado, Interação humana pode ser um tema complexo, e possui diversas abordagens. Cabe a nós, entender o lado positivo deste investimento para a humanização das máquinas e compreender que ainda temos algo que nos difere de qualquer outro ser existente, a nossa consciência. E você, já parou pra pensar nas tarefas simples que você fazia antes, e que agora pertencem a uma máquina?

Interação humana e a evolução das máquinas.

22

interação-humana-e-a-evolução-das-máquinas-1213c22b8116

2018-06-11

2018-06-11 15:27:28

https://medium.com/s/story/interação-humana-e-a-evolução-das-máquinas-1213c22b8116

false

1,201

null

Interacao

interacao

Interacao

19

Paulo Ricardo

Apaixonado por tecnologia e todas as formas de comunicação e aprendiz assíduo de conteúdos de UX Design, tipografia e Interfaces.

121f506ae6b7

paulo_ricardo

112

244

20,181,104

null

0

null

0

null

2018-07-17

2018-07-17 04:53:43

2018-07-17

2018-07-17 05:05:16

1

false

en

2018-07-17

2018-07-17 05:07:41

2

121410005bac

1.792453

0

Artificial Intelligence is simplifying every human activity today. With self-driven cars, automated purchase systems, and chatbots, AI is…

5

How Is Artificial Intelligence Stepping Up The E-Commerce Game? Artificial Intelligence is simplifying every human activity today. With self-driven cars, automated purchase systems, and chatbots, AI is proving to become a race of its own. AI and e-commerce, though in no way related, are two trendy topics. Almost all businesses are going digital, and the one question left unanswered is — is AI here to help? Artificial Intelligence Here are 4 powerful AI implementations for your e-commerce business: 1) Customer-Centric Searches: A huge let down for e-commerce customers is when they search for a product but the results show something else. Having AI-powered searches will enable the system to evaluate user searches and products to deliver the most relevant output to the customer. With machine learning and artificial intelligence, you as an owner will not have to manually customize anything; the system will do it all! 2) Retarget to Retain: Customers like the “special treatment”; and as an e-commerce business owner, it is very easy to deliver this. Using AI, customer search data can be collected and the next time a customer logs in, give them discounts/offers based on products they searched! By doing this you are retargeting potential buyers and retaining them to make them a 100% confirmed buyer. 3) Automated Recommendations: A slight fraction of the customer circle might find it annoying, but the majority of online shoppers love recommendations. It makes shopping times convenient! Looking for a pant and seeing belt suggestion improves the chance of the customer buying either; alternatively, selling them as a combo might get the customer to buy the entire package! AI offers automated recommendations that will multiply customer growth. 4) Chatbots: Having to talk to 20–30 customers consecutively is easy. Owning an e-commerce site with 300–400 live visitors all asking queries at the same time makes tending to each one tough. AI offers an amazing solution to this — chatbots. Chatbots are designed to recognize keywords and respond to user queries in a human-like manner! Artificial Intelligence strives day in and day out to simplify the user experience and enhance the several functionalities to achieve human-like or even higher levels of excellence. If you own an e-commerce business and want to make it more efficient, AI is the perfect go-to. At Openwave, our developers are geared head to toe in e-commerce app development and integration. Give your business the leverage it needs; contact our app development team and avail stellar AI services.

How Is Artificial Intelligence Stepping Up The E-Commerce Game?

0

how-artificial-intelligence-is-stepping-up-the-e-commerce-game-121410005bac

2018-07-17

2018-07-17 05:07:41

https://medium.com/s/story/how-artificial-intelligence-is-stepping-up-the-e-commerce-game-121410005bac

false

422

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Melanie Antoine

Business Development Executive at Openwave Computing LLC. Compulsive blogger, mobile junkie and app security expert too. My areas of interest are mobile app dev

23c9e598879b

OpenwaveComputing

309

1,287

20,181,104

null

0

null

0

6ce2e399e529

2018-06-28

2018-06-28 09:27:16

2018-07-02

2018-07-02 13:38:14

1

false

en

2018-07-02

2018-07-02 13:38:14

2

12153eaf7971

4.611321

10

0

Critical business issues are being automated and solved by AI. Companies are extracting more value than ever before from the data they…

5

Simulation: the bedrock of AI Critical business issues are being automated and solved by AI. Companies are extracting more value than ever before from the data they collect. And these trends look set to continue. But AI is still struggling to help business leaders in industries like finance, transportation and healthcare make better decisions in the complex systems they operate in. Andrew Ng — one of the leading figures in the world of AI and Machine Learning — explains how the leading companies in the AI space are comitting huge amounts of resource to generate the data they need to train their intelligence on. Uber, Google and Tesla, for example, are spending billions of dollars covering millions of kilometers in their self-driving cars to perfect the algorithms that drive them. They, along with many others, are leveraging an alternative way of generating accurate data to train these algorithms, and it’s a technology that has been around for a number of years. Computational Simulation. In a recent episode of the McKinsey & Co. podcast, McKinsey Global Institute partner Michael Chui explains the role that simulation plays in training AI: “To try to improve the speed at which you can learn some of those things, one of the things you can do is simulate environments. By creating these virtual environments — basically within a data center, basically within a computer — you can run a whole bunch more trials and learn a whole bunch more things through simulation. So, when you actually end up in the physical world, you’ve come to the physical world with your AI already having learned a bunch of things in simulation.” Simulation complements Artificial Intelligence. To more deeply understand why this is, we can think of machine learning and simulation as being two different approaches for understanding and predicting the behaviour of complex adaptive systems. Individually they are both powerful modelling paradigms, but it is together, working in concert, that they show the greatest promise. Machine learning automates the building of analytical models. The models are constructed using algorithms that learn from data. These algorithms are able to update models in real-time, giving them the ability to generate real-time predictions. What’s more, they are able to learn from past predictions, outcomes, and errors. This makes them incredibly powerful. But these models are limited to forecasting effects of events that are similar to what has already happened in the past. The models are likely to produce inaccurate results once they extrapolate beyond previously observed bounds. If the future will look different from the past, if we have missing data, or if there is a bias in our data, then data-driven methods are considerably flawed. Many bankers found this out to their considerable cost when the financial crisis hit in 2008. An over-reliance on data-driven models for decision-making in complex systems is risky; their key deficiency is the scarcity of observations during crisis times. This makes them fragile under the definition prescribed by Nicholas Taleb: “Anything that has more upside than downside from random events (or certain shocks) is antifragile; the reverse is fragile.” In other words, when exposed to previously unseen events, data-driven models tend to do worse than when history is repeating itself. Simulation’s key advantage over data-driven methods is that it allows us to forecast things that have never happened before and to run scenarios outside of historical bounds — including crisis scenarios. This is no panacea. The caveat is that we need a good theory and causal hypotheses about how the system we are studying works. Simulation works best when the processes of the system under study are well-understood such that high-fidelity simulations are able to match predicted output. As long as our theory is sound, we can make startling accurate predictions about states of the world we have never seen before. When reality pans out along one of these simulated trajectories, we already know what to do because we have already test-driven our decisions in the virtual world. Machines learning from simulations Probably the most familiar example of simulation is the flight simulator. Pilots expose themselves to previously unseen states of the world in a realistic model of flight in order that they may make better decisions when flying in the real world. Just as pilots are ‘trained’, so too are machine learning models. We talk about ‘training’ them on data, just as the pilot’s brain is learning from the large amounts of visual and sensory data being generated by the flight simulator. Unleashing machine learning models on accurate simulations is similarly powerful. Rather than have that model fail when reality moves outside the known parameter set, we can generate vast quantities of data using our simulation — even training the model in the extreme tails of the distribution — in order that our data-driven models can provide useful insight during crisis times. Simulations learning from machines We can also use machine learning to build better simulations. Recall that in order to provide sensible predictions, simulations need to accurately reflect the processes at work in the system of study. We can use machine learning here to inform behaviours in our simulation models. Less well-understood processes can be approximated by machine learning models themselves; where a behavioural component of a simulation model is hard to pin-down, we can use data-driven black box functions to stand-in for stronger theoretical foundations. This type of ‘hybrid’ modelling — mixing different modelling paradigms — allows us to alloy them into something more powerful. Machines adapting to a changing environment The most challenging aspect of making decisions in a complex adaptive system is their adaptive nature. Any human knows that the actors, or agents, operating in these systems are not fixed and unresponsive. Instead, they change their behaviour as the incentives set by their environment change. Agent-based simulation models are even able to account for the ability of agents to deviate from rationality, optimise, or exploit their environment. Given a policy change, these models can predict how a system will reconfigure itself — allowing machine intelligence to stay ahead of the human responses. One real-world example of how this can be applied is detecting fraud. We can train machine intelligence for spotting fraud on a simulation comprised of “intelligent fraudsters” which are able to learn fraudulent strategies that humans may not have yet-devised in the real-world. It is not a question of simulation or machine learning, but the combination of simulation and machine learning that offers the greatest potential. Simulation is a foundational technology for businesses wanting to exploit AI. The technology to build powerful simulations already exists. And rather than collecting data in the real-world, more and more companies are turning to virtual worlds to collect lifetimes’ worth of data for a fraction of the cost. The winners in the race to build AI will be those companies best able to leverage simulation to train them.

Simulation: the bedrock of AI

127

simulation-the-bedrock-of-ai-12153eaf7971

2018-07-03

2018-07-03 07:43:35

https://medium.com/s/story/simulation-the-bedrock-of-ai-12153eaf7971

false

1,169

We founded Simudyne with the goal of improving decision making across both private and public sectors by leveraging advances in computational power and complexity science. Simudyne trains intelligence in high-fidelity simulated environments to make better decisions.

null

Simudyne

support@simudyne.com

simudyne

AGENT BASED MODELING,FINANCE,SIMULATION,TECH,AI

simudyne

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

John Hill

Economist & Computer Scientist. Interested in complex systems, modelling & simulation. Simulations Engineer at Simudyne.

31e90b628779

john89hill

29

22

20,181,104

null

0

null

0

null

2017-11-08

2017-11-08 00:36:24

2017-11-22

2017-11-22 23:22:32

5

false

en

2017-11-22

2017-11-22 23:22:32

0

12169e819d27

3.380503

12

0

Machine learning, AI, Deep Learning are trending buzzwords now. Especially with Site Search and Online Commerce.

5

Machine Learning v/s Automation in Site Search for Ecommerce Machine learning, AI, Deep Learning are trending buzzwords now. Especially with Site Search and Online Commerce. How do you tell that you need machine learning? How do you know that a particular solution has machine learning? Sometimes automation and machine learning are used interchangeably. The biggest difference between the two is that automation does not require any ‘learning’. Automation is all about using machines to automate what humans would otherwise do manually. For instance in E-commerce Site-Search, a good automation based solution will use past purchase information to rank products with high sales numbers at the top. It might even use ‘clicks’ to automate showing highly viewed products to the top. This is very useful because many search systems today require humans to input rules to boost products to the top. Popularity Automation in action: Leggings are showing up in pants search results The flip side of automation is that it can go wrong with ranking within search results(see fig 1 above). If a ‘legging’ is selling really well, it’s great if it shows up when somebody searches for ‘pants’. The system is geared towards ranking the most popular products to be shown the most. What about 80%+ visitors who actually want to see pants, not just the most popular legging(which technically is also a pant). This ends up catering the top 20% of best selling products to 20% of visitors shopping online. So what is Machine Learning? If you need a system that not only automates human tasks but also sifts through a lot of data to identify patterns of predictability in data sets, you need Machine Learning. Machine Learning systems require intensive computations (a lot of CPUs) and need training data sets (a lot of data). They will use that to predict for instance the likelihood of a user to click on a certain type of product after a certain search. Another use case would be to predict what other queries are similar to a particular query with different search results. The limiting factor for machine learning is the data — the more data it ingests, the more likely results will be realistic. Machine learning models cannot predict outside of the data that is fed in. This is an important thing — If somebody is offering machine learning and is not ingesting a HUGE amount of data, they are just automating, there is no machine learning in play. A typical use case of machine learning when it ingests source of traffic, search queries, clicks on filters, click rank of products, product clicks, time spent on PDP pages, add-to-cart events, purchases, exits, repeat visits, non search clickstream events and web bases social trending data, it is likely to be trained well to perform. Assuming relevancy works well with leggings and pants categorized under ‘pants’ it predictively ranks products differently for ‘pants’ and ‘leggings’ With machine learning, while relevancy plays its part, it is able to determine that there is sufficient deviation in behavior when people search for ‘pants’ v/s ‘leggings’ on its own. It’s not programmed or automated. See fig 2 above. At Unbxd, when we compared a machine learning based search system (CTRs and conversion rates) on an A/B test with the best automated systems, they both behaved similarly on the top 5% of queries. However, the MC based search outperformed automated search by 2x (100% difference in conversions) with the rest 95%.This is because the top 5% search queries are driven by heavily by sales data. The remaining 95% queries need more predictability — not just automation. One thing to note though is to be realistic about what machine learning can deliver. You need significant data to make training models work. If your site has less traffic — it might take a while before it starts kicking in. To summarize you can use this simple checklist to verify if there a site search has machine learning or not: Does site search ingest at-least 5 different types of signals? What machine learning models does it use? What are some patterns that these models have predicted? Is there a dedicated data sciences team? What’s the technical architecture for machine learning in play?

Machine Learning v/s Automation in Site Search for Ecommerce

95

machine-learning-v-s-automation-in-site-search-for-ecommerce-12169e819d27

2018-06-17

2018-06-17 11:35:13

https://medium.com/s/story/machine-learning-v-s-automation-in-site-search-for-ecommerce-12169e819d27

false

675

null

Machine Learning

machine-learning

Machine Learning

51,320

pavansondur

CEO @ Unbxd. Passionate to solve hard technology problems.

aa914e8892f4

pavansondur

71

130

20,181,104

null

0

null

0

null

2018-08-30

2018-08-30 16:19:03

2018-08-30

2018-08-30 16:19:18

0

false

en

2018-08-30

2018-08-30 16:19:18

1

1216f4642de7

1.622642

0

Download ebook Grokking Deep Learning PDF EPUB KINDLE By Andrew W. Trask Link https://shoppipubherenow.icu/?q=Grokking+Deep+Learning…

1

DOWNLOAD PDF eBook Free Grokking Deep Learning By Andrew W. Trask PDF Books #Audiobook Download ebook Grokking Deep Learning PDF EPUB KINDLE By Andrew W. Trask Link https://shoppipubherenow.icu/?q=Grokking+Deep+Learning Artificial Intelligence is one of the most exciting technologies of the century, and Deep Learning is in many ways the ?brain? behind some of the world?s smartest Artificial Intelligence systems out there. Loosely based on neuron behavior inside of human brains, these systems are rapidly catching up with the intelligence of their human creators, defeating the world champion Go player, achieving superhuman performance on video games, driving cars, translating languages, and sometimes even helping law enforcement fight crime. Deep Learning is a revolution that is changing every industry across the globe.Grokking Deep Learning is the perfect place to begin your deep learning journey. Rather than just learn the ?black box? API of some library or framework, you will actually understand how to build these algorithms completely from scratch. You will understand how Deep Learning is able to learn at levels greater than humans. You will be able to understand the ?brain? behind . . . . . . . . . . . . . . Read Online PDF Grokking Deep Learning, Download PDF Grokking Deep Learning, Download Full PDF Grokking Deep Learning, Download PDF and EPUB Grokking Deep Learning, Read PDF ePub Mobi Grokking Deep Learning, Reading PDF Grokking Deep Learning, Read Book PDF Grokking Deep Learning, Read online Grokking Deep Learning, Download Grokking Deep Learning Andrew W. Trask pdf, Download Andrew W. Trask epub Grokking Deep Learning, Read pdf Andrew W. Trask Grokking Deep Learning, Download Andrew W. Trask ebook Grokking Deep Learning, Read pdf Grokking Deep Learning, Grokking Deep Learning Online Download Best Book Online Grokking Deep Learning, Read Online Grokking Deep Learning Book, Read Online Grokking Deep Learning E-Books, Read Grokking Deep Learning Online, Read Best Book Grokking Deep Learning Online, Read Grokking Deep Learning Books Online Download Grokking Deep Learning Full Collection, Download Grokking Deep Learning Book, Read Grokking Deep Learning Ebook Grokking Deep Learning PDF Read online, Grokking Deep Learning pdf Download online, Grokking Deep Learning Read, Download Grokking Deep Learning Full PDF, Read Grokking Deep Learning PDF Online, Read Grokking Deep Learning Books Online, Read Grokking Deep Learning Full Popular PDF, PDF Grokking Deep Learning Read Book PDF Grokking Deep Learning, Read online PDF Grokking Deep Learning, Download Best Book Grokking Deep Learning, Read PDF Grokking Deep Learning Collection, Read PDF Grokking Deep Learning Full Online, Read Best Book Online Grokking Deep Learning, Download Grokking Deep Learning PDF files #pdf #pdfdownload #BookOnline #WORD #Mobipocket

DOWNLOAD PDF eBook Free Grokking Deep Learning By Andrew W. Trask PDF Books #Audiobook

0

download-pdf-ebook-free-grokking-deep-learning-by-andrew-w-trask-pdf-books-audiobook-1216f4642de7

2018-08-30

2018-08-30 16:19:18

https://medium.com/s/story/download-pdf-ebook-free-grokking-deep-learning-by-andrew-w-trask-pdf-books-audiobook-1216f4642de7

false

430

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Cailyn Simon

null

71dcbd448351

cailynsimon

0

1

20,181,104

null

0

null

0

6c35dc0cba4d

2017-09-28

2017-09-28 05:53:05

2017-10-31

2017-10-31 10:26:00

4

false

en

2017-10-31

2017-10-31 10:26:00

3

121857434837

2.5

37

3

0

Applying machine learning to the classifieds experience.

4

Deep image understanding at Carousell In the past five years, Carousell has led the way in mobile classifieds and is one of the fastest growing mobile marketplaces in Southeast Asia. We’re in 19 cities across 7 countries. We’re looking at ways to leverage machine learning to enhance the user experience. At Carousell, my team develops machine learning features that help our users list, sell, and buy items more easily. We train our models on Carousell’s sizeable internal datasets of items for sale and user interactions. Our first feature powered by machine learning suggests titles and categories for your listing, based on the images that you upload. This is available in Singapore on the Android app, and is in the process of rolling out on iOS and in other countries. Suggested categories and titles based on the image. The network’s third suggestion “Yamaha Keyboard” is correct. We train deep convolutional neural networks on our database of tens of millions of listings, to classify images into their categories. This classifier is used to provide category suggestions in the app. However, treating title prediction as a categorisation task like this would not work well, as there are so many different titles in our data. Instead, we trained a ranking model that takes an image and attempts to select the correct title out of a pool of candidate titles. The neural network for ranking titles has two halves. One half looks at the image using deep convolutional layers; the other looks at potential titles, processing the words and phrases using embeddings and a deep neural structure. The two halves map images and titles to a shared high-dimensional vector space, and vector similarity is then used for ranking. Our network is learned jointly from scratch with a single ranking loss function. This structure allows for a lot of pre-computation in training and inference. When a new image is uploaded to Carousell, the model ranks a list of titles derived from hundreds of thousands of listings to find good suggestions in under 100 milliseconds. The shared image and title space learned by the deep neural network. The network has learned to put images and their corresponding titles nearby. It has learned implicit clusters like clothes, games and electronics. It still makes some mistakes, for example it put the title “IKEA cushion” too close to the image of the Hermes handbag, and it did not learn to identify the “Sketch Drawing” with high confidence. The high-dimensional space is projected down to 2 dimensions for the visualisation. The difference between the deep vector representation of the red phone case and the grey one gives a semantic ‘red’ direction in the vector space. Adding the red vector to other images allows us to ‘turn them red’. We train our models across multiple GPU machines in parallel for hundreds of millions of steps (but keep training time down to a couple of days to allow for quick development). Our best network is a joint model that predicts the category and ranks titles using a shared deep representation of the image. A larger sample of the vector space learned by the network, showing only images. Some well-defined clusters include women’s shoes at the top, clothes at the bottom, and mobile phones to the left. If you’re already using these new machine-learning powered features on our marketplace, thank you for trying them. If you haven’t yet, we hope you’ll give it a try soon. We expect the real learning to happen from your interaction with the features, i.e. which suggestions you click on, and as more and more people use it. We are currently hiring data scientists and machine learning engineers to join us in building more features like this. Carousell / Careers Inspiring everyone in the world to start selling At Carousell, we want to make selling simple for everyone. We all have…careers.carousell.com

Deep image understanding at Carousell

352

deep-image-understanding-at-carousell-121857434837

2018-04-02

2018-04-02 00:32:21

https://medium.com/s/story/deep-image-understanding-at-carousell-121857434837

false

477

What's going on under the hood at one of the world's largest and fastest growing mobile classifieds marketplace. We're on a mission to inspire everyone in the world to start selling.

null

carousell.sg

null

Carousell Insider

null

carousell-insider

STARTUP LESSONS,HR,ENTREPREURSHIP,DESIGN,TECH

null

Machine Learning

machine-learning

Machine Learning

51,320

Matt Henderson

machine learning for natural language understanding

a57d7dfb15df

matthen

419

209

20,181,104

null

0

null

0

null

2018-08-07

2018-08-07 03:50:25

2018-08-07

2018-08-07 04:17:55

7

false

en

2018-08-07

2018-08-07 04:17:55

9

12190174a1d3

2.933019

1

0

Hello to the Bottos Community Discovery Community wants to take a second to thank all our supporters we really appreciate all the love and…

5

How to vote for Discovery Community Node — Bottos AI Node candidate Hello to the Bottos Community Discovery Community wants to take a second to thank all our supporters we really appreciate all the love and will be striving to be the best Node for Bottos chain and the Bottos community! As the voting period is coming up 12:00:00, 8/8 to 12:00:00 8/12 (UTC time) We wish all other candidates good luck and we hope the Bottos Community votes for us! Our ERC20 Address is 0x0bdb2d0251418054c87f2c8b6f5320973ff94ce2 You can also find it on https://node.bottos.org/index.php/index/election/sort/thumbs_up/lang/en.html login to mew adding custom token For the BottosVote Token Discovery Node will guide you on how to add custom token and to get voting! After you connect to MEW (MyEtherwallet *PLEASE CHECK URL AND CERTIFICATE AVOID PHISHING*) whether you are using hardware wallet recommended, click the “Add Custom Token” box on the right side of MEW. Information for Custom token Contract: 0x17303800fec5a38f6070c3356445ca3e41e8a173 Decimals: 0 Symbol: BottosVote Enter all token information into text field and click the “Save” Button. Now you should see your balance of BottosVote Token under your custom token tab. How to vote for candidates How to send and cast your vote for candidate is going to Node campaign Our 100% community-based team is represented by Shreder, admin of the Western Telegram group, and Bottos core member…node.bottos.org And getting the address to vote, If you’re voting for Discovery Community Node you send your BottosVote Tokens to this ERC20 address 0x0bdb2d0251418054c87f2c8b6f5320973ff94ce2 Remember to only send BottosVote Tokens, DO NOT SEND BTO/DTO! Generate and confirm transactions After you have correctly entered the ERC20 Address 0x0bdb2d0251418054c87f2c8b6f5320973ff94ce2, you Generate and confirm your transaction. And you have officially supported Discovery Node! Please beware of some basic rules and time frame set by the foundations to avoid your votes being invalidated! Will Discovery Community Node give addition rewards? Because Discovery is a community-based team, we believe that it is only fair that our supporters receive DTO airdrops directly. As a thank you for supporting us and the Bottos AI Ecosystem we will allocate 5% of our Block Production rewards for distribution to our voters. Additionally, we plan to consider other incentives to give back to the Bottos community and our supporters in the future. 1 BottosVote Token = 10 DTO First come, First serve basis. We will take remaining DTO from our 5% pool if any and create lottery Airdrops. The more we earn the more the community gets! What else will Discovery Community do for the Bottos Community? In addition to supporting DPOS blockchain consensus, Discovery Community Node will provide a platform for discussions, tutorials, and engagement among Bottos supporters. Input from the community will be considered in all governance decisions. We want to create a hub where the community of Bottos can be heard! Discovery Node ready for launch! Discovery Community Node is excited to participate in this campaign and would greatly appreciate your vote and support! Discover and Explore the Bottos New AI Ecosystem with us! Our Social Media Website- http://bottoscommunity.com Twitter- https://twitter.com/DiscoveryNode Telegram- https://t.me/DiscoveryNode Medium- https://medium.com/@DiscoveryNode Reddit- https://www.reddit.com/r/DiscoveryNode Steemit- https://steemit.com/@discoverynode Youtube- https://www.youtube.com/channel/UCTFdMPq0hzxIYTkB3zaE57g

How to vote for Discovery Community Node — Bottos AI Node candidate

17

how-to-vote-for-discovery-community-node-bottos-ai-node-candidate-12190174a1d3

2018-08-07

2018-08-07 04:17:55

https://medium.com/s/story/how-to-vote-for-discovery-community-node-bottos-ai-node-candidate-12190174a1d3

false

499

null

Blockchain

blockchain

Blockchain

265,164

DiscoveryNode

Bottos Block producing candidate actively engaged with Bottos community around the globe and promoting Bottos new AI Ecosystem innovating AI + Blockchain

b16d14bff573

DiscoveryNode

7

2

20,181,104

null

0

null

0

null

2018-08-15

2018-08-15 17:53:49

2018-08-15

2018-08-15 18:10:10

2

false

tr

2018-08-15

2018-08-15 18:10:10

0

121941532f5c

2.511635

12

0

MATRIX AI Network’ün hem Bitfinex hem de Ethfinex borsalarında listelendiği için güven yüksek! Daha fazla likidite ve fiat eşleştirmeleri…

5

MATRIX AI Network’ün Bitfinex’te listelenmesinin pozitif yönleri neler olacaktır? MATRIX AI Network’ün hem Bitfinex hem de Ethfinex borsalarında listelendiği için güven yüksek! Daha fazla likidite ve fiat eşleştirmeleri ile, MATRIX topluluğuna katılmak her zamankinden daha kolay! Matrix AI Network , hem Bitfinex hem de Ethfinex borsalarında 09–08–2018 tarihi itibariyle resmi olarak listelenmiştir. Özellikle, 2012 yılında kurulan ve dünyanın en büyük borsalarından biri olan Bitfinex’in , Matrix yönetiminin ve stratejik yatırımcıların en önemli önceliği olduğu dikkat çekmektedir. Bu borsalarda listelenme, hem uluslararası pazarda hem de Çin’de blockchain topluluğuyla yüksek katılımları destekleyeci bir özelliktedir. Temmuz ayı sonlarında ;Çin’de, Matrix Baş Yapay Zeka Bilim adamı Steve Deng ile blockchain ile ilgili önde gelen bir çok kişiye röportaj verdi. Matrix CTO ve Baş Ağ Mimarı Dr. Bill Li, Çin’de bilgi işlem için Ulusal Standartlar Formülasyon konferansında ayrıca Blockchain ve dağıtılmış muhasebe teknolojileri konusunda Kunning’de bilgilendirme yaptı. Matrix CEO’su Mr. Owen Tao ve Matrix Çip Şef bilim adamı Dr. Tim Shi ise İstanbul’u ziyaret ederek bilgilendirmelerini gerçekleştirdiler. Matrix AI Network İş Yöneticisi olan Mr. Simon Han’a göre, Bitfinex’de listelenmek , toplumun Matrix’i, daha güçlü ve kapsamlı küresel bir Blockchain projesi olarak kabul etmesini sağlayacaktır” olarak ifade etmiştir. MATRIX gibi bir proje için büyük borsalarda listelenmesinin önemi, ekibinin teknik gücü ya da projenin potansiyeli ne olursa olsun, topluluk tarafından benimsenmesi, başarı için en önemli unsurdur. Blockchain projeleri topluluk desteği olmadan hiçbir şekilde ilerleyemez ve hayatta kalamaz. Tarih, likidite eksikliğinden , popüler borsa listelenmesi ve topluluk desteği olmayan sayısız başarısız projelerle doludur. Bu nedenle, toplumun gözünden önemli yapıtaşlarına bakabilmek için zamana ihtiyaç vardır. Matrix’in Bitfinex’de listelenmesi, topluluğu için nasıl bir fayda sağlayacak? Kolay Benimseme MATRIX AI Network topluluğu güçlü olarak büyüyor ve topluluğa her gün daha fazla kişi katılıyor. Engelleri ve zorlukları atlatmak için en önemli nokta, kolay benimseme ile mümkün olacaktır. Bitfinex gibi büyük borsalar toplulukla doğrudan etkileşime girdiğinden dolayı önemi daha da artmaktadır. Trader’lar borsaya girişlerinde 4–15 coin arasında birden fazla işlem yaptıklarından dolayı çok fazla zaman harcamaktadırlar. Bu nedenle bir üst onay almaya yönelik engelleri azaltarak, birden fazla şifreleme ile borsalarda listelenmek, yeni benimseme seviyelerine ve piyasa etkisine olumlu etki yapabilecektir. Fiat Desteği Fiat para birimleri için destek de üst borsaların önemli bir özelliğidir. Fiat para birimi desteği olmadan, trader’lar, diğer “geçitli” olan borsalardan BTC veya ETH satın almak zorunda kalırlar. Ancak bu şekilde diğer borsalara coinlerini aktarılabilir. Tabi gönderim yapılırken birçok masrafı ile birlikte! Neyse ki, Bitfinex bu sorunu önlemek için, bir fiili eşleştirmeyi (USD / MAN eşleştirmesi içermekte) ve doğrudan fiat para çekme işlemlerini de kolaylaştırarak üyelerine sunmaktadır. Modern en ileri Ticaret Bitfinex’in deneyimli trader’lara hitap eden bir takım araçları vardır. Son Encrybit araştırmasına göre, Bitfinex, yerinde teknik analiz ve algoritmik ticaret seçenekleri için borsalar arasında en iyi UI’ye sahip olanıdır. Bir şamdan şeması gibi temel özelliklerinin yanında Fibonacci araçları , trend çizgi ve göstergeleri gibi özellikleri mevcuttur. Kar ticareti, kısa satış ve boşluk doldurma, öldürme, iceberg, OCO, durdurma emirleri ve emir gönderme gibi araçları olduğu için ticaret yapmaya yarayan tüm unsurları barındırmakta olduğundan, borsa araçları ile ilgili sıkıntısı bulunmamaktadır. Bitfinex ve Ethfinex borsalarına giriş, hem MATRIX projesinin hem de MATRIX’in tutkulu topluluğunun bir zaferidir. MATRIX kesinlikle topluluk desteği olmadan bu başarıyı yakalayamazdı. Topluluk desteği kesinlikle MATRIX AI Network’ün geleceğini de belirleyecektir! Mr. Simon Han son olarak şu sonuca varıyor; “Bitfinex borsası, gelecekteki yatırımcılar ve ortakların, MATRIX ekosistemine katılmaları için gittikçe büyüyen bir güven temelini oluşturuyor. Dünyanın en iyi topluluk blockchain’i olma hedefimize birlikte ulaşacağız. ”

MATRIX AI Network’ün Bitfinex’te listelenmesinin pozitif yönleri neler olacaktır?

520

matrix-ai-networkün-bitfinex-te-listelenmesinin-pozitif-yönleri-neler-olacaktır-121941532f5c

2018-08-15

2018-08-15 18:10:11

https://medium.com/s/story/matrix-ai-networkün-bitfinex-te-listelenmesinin-pozitif-yönleri-neler-olacaktır-121941532f5c

false

564

null

Blockchain

blockchain

Blockchain

265,164

Matrix AI Network Turkey

Official Turkish Community

c0d52c47b03d

matrixaiturkey

96

415

20,181,104

null

0

null

0

null

2018-08-27

2018-08-27 15:45:12

2018-08-28

2018-08-28 06:09:38

5

false

en

2018-08-28

2018-08-28 06:09:38

12

121a501645e5

2.889937

0

An intense and hands-on deep tech journey, a three-day coordinated transfer of knowledge and an excellent and unique way to explore the…

5

Speakers from Amazon and Cisco join Codiax 2018 | Why you can’t miss the 2nd edition An intense and hands-on deep tech journey, a three-day coordinated transfer of knowledge and an excellent and unique way to explore the deep technology innovations happening now. This is Codiax! Deep Tech is coming to Eastern Europe — November 15–16, Cluj-Napoca and November 17, Sibiu Focused on the technologies of the future and the ways they impact our lives it is specially designed for senior software engineers, data scientists, CTO’s, VP’s of Engineering, founders and software architects who dare to expand their know-how and discover new boundaries. We bring the latest and best content and showcases of real-life applications. To make sure this happens in a closely-knit environment we have limited the participation to 350 attendees. During these three days we will decipher five hot topics from all angles, trying to open new perspectives for the attendees: Artificial Intelligence, Virtual Reality, Augmented Reality, Blockchain, Internet of Things and Frontier Hardware. One game changer topic in the deep tech ecosystem is for sure AI and its global impact. AI could contribute up to $15.7 trillion to the global economy by 2030, more than the current output of China and India combined. Of this, $6.6 trillion is likely to come from increased productivity and $9.1 trillion is likely to come from consumption-side effects. All this according to PwC Consumer Intelligence Series: Bot.Me, 2017. 67% executives say AI will help humans and machines work together to be stronger using both artificial and human intelligence To explore, in depth, the impact and the power of AI and how it can be used and developed we are delighted to recommend the talks and workshops of our top-notch speakers for the 2nd edition of Codiax. Julien Simon, Principal Technical Evangelist at Amazon Daniel Wroblewski, Head of Engineering at EQT If you combine AI with machine learning you will be amazed of how many helpful real-world examples you get. For example, Siri and Cortana use machine learning and deep neural networks to imitate human interaction, PayPal uses ML algorithms to combat fraud, Netflix’s video recommendation engine is powered by machine learning and Uber uses ML algorithms to determine arrival times. Not only in these services, but also in the manufacturing ML will play a decisive role in the next few years. McKinsey predicts machine learning will reduce costs related to transport and warehousing and supply chain administration by 5 to 10% and 25 to 40%, respectively. Due to machine learning, overall inventory reductions of 20 to 50% are possible. Take a closer look at what Machine Learning is bringing to the future with leaders and experts from the industry: Edwin Witvoet, Co-Founder of Spyhce Enzo Fenoglio, Data Scientist at Cisco In 2017, participants from all over Europe gathered at Codiax to experience the workshops, the truly inspirational speakers and the atmosphere. This year make sure you are one of them! Our previous speakers include: Alexey Ershov (VP at IBM Watson IoT), Vlad Lata (CTO & Co-Founder at KONUX), Sergii Kharagorgiiev (Chief Computer Vision Engineer at Starship Technologies), Mario Alemi (Scientist at GetJenny), Kryztof Urban (Data Science Manager at GumGum), , Camilo Martinez (Software Development Team Lead at Booking.com), Newton Howard (Director of the MIT Synthetic Intelligence Lab) and others Event updates can be found on our website, Facebook, Twitter and Instagram.

Speakers from Amazon and Cisco join Codiax 2018 | Why you can’t miss the 2nd edition

0

speakers-from-amazon-and-cisco-join-codiax-2018-why-you-cant-miss-the-2nd-edition-121a501645e5

2018-08-28

2018-08-28 06:09:38

https://medium.com/s/story/speakers-from-amazon-and-cisco-join-codiax-2018-why-you-cant-miss-the-2nd-edition-121a501645e5

false

545

null

Machine Learning

machine-learning

Machine Learning

51,320

CODIAX

null

b8d253988e10

codiaxconf

2

3

20,181,104

null

0

null

0

null

2018-03-09

2018-03-09 20:26:12

2018-03-09

2018-03-09 20:26:44

2

false

en

2018-03-09

2018-03-09 20:26:44

35

121acfbd0bfc

3.349371

2

0

Technology is moving fast. You may feel like you’re having a hard time keeping up. Deep learning neural networks? What’s a TensorFlow…

3

How to get your learn on. Technology is moving fast. You may feel like you’re having a hard time keeping up. Deep learning neural networks? What’s a TensorFlow? Serverless event-driven architectures? Distributed acyclic graphs?! Luckily there are lots of ways to learn. That learning will take you from being a “HA HA BUSINESS” person to “I’m from the internet and I’m here to help.” Websites Just bookmark them under a tab called “Technology” and visit them when you’re having your morning coffee. Here’s are some of my favorites: Trending tech stacks: Stackshare.io, Analyzo, StateofDev General technology sites: Wired, TechCrunch, Gigaom Company blogs: Facebook, Google, Uber Artificial intelligence: Distill.pub Best visual essays and data science: The Pudding Twitter Twitter is like a stream of information that you can dip into throughout the day. Start by following a few leaders in the fields and then look at who they follow. Rinse, repeat. It’s a great way to get exposed to upcoming events and shared articles. (You can start by following me https://twitter.com/dstepp2) Medium The best blog platform — I mean look, even freaking Netflix has their own blog on here. Like Twitter, follow leaders in your field, topics you’re interested in, and customize it based on their suggestion algorithm. My two favorites that post lots of content: General technology: https://hackernoon.com/ Cloud computing: https://read.acloud.guru/ Online Courses On demand training in anything you want — by the absolute best teachers in that field. Learning from experts is the most powerful tool in your learning belt. With price points as low as $10, you can’t afford to not take training! Just set aside 20 minutes a day to do a little at a time. Best value and interface. Wait for their monthly sales and load up with a lot of $10–15 courses: https://www.udemy.com/ Real screen grab from my Udemy dashboard. I can’t help myself buying more courses. Get trained on GCP and Azure: https://cloudacademy.com/ Get trained on AWS and Serverless: https://acloud.guru/ There are also tons of free college level courses. Just do a Google search and find ones that interest you. Youtube Seriously, just follow Siraj Rival. Siraj posts about 3 videos each week on artificial intelligence and machine learning. He breaks down extremely complicated topics into easily digestible pieces. And he’s fun to watch! ACloud.Guru also produces a lot of great content each week. I really enjoy their weekly AWS, and monthly Azure roundups. It’s an easy way to make sure I’m on top of all the new cloud releases. Reddit and Hackernews Don’t install the Reddit app on your phone if you ever want to have free time again. (My wife gets on me about being on my phone too much. It’s that damn Redditisfun app’s fault!) Hackernews is filled with really smart people posting about technology. Reddit is filled with all types of people, so be cautious. Here’s a few subs to get you started: https://www.reddit.com/r/artificial/ https://www.reddit.com/r/bigdata/ https://www.reddit.com/r/MachineLearning/ https://www.reddit.com/r/programming/ https://www.reddit.com/r/technology/ Podcasts Podcasts are my favorite way of learning about new topics while on the move. I have episodes loaded up on my phone and listen to them whenever I’m driving or commuting on the train. Sometimes I’ll go wild and listen to one while cooking dinner. Multitasking for the win! Here are some of my favorites: AI in Industry GCP Podcast Software Engineering Radio QnA Sites There is a joke that the only qualifying skill you need as a developer is the ability to search Stackoverflow. That’s because it pretty much contains the answer to every single development question ever. If you can’t find the information here, godspeed and good luck. Quora is another place I check for some general knowledge questions; or I’ll even check out AlternativeTo for new software recommendations. Good old fashion books You know, the things where information was originally stored. I’m currently reading The DevOps Handbook (I’m not too impressed yet. Having just seen Gene Kim speak in person, I have to say the book is a bit dated. Spends too much time talking about how great DevOps is without much practical advice…) I have a long reading list on my Kindle (much like Podcasts and Udemy courses), including the upcoming book from Accenture CTO Paul Daughtery, Human + Machine: Reimaging Work in the Age of AI. I try to get at least an hour of time each day dedicated to learning new information — from podcasts during my commute, videos during lunch, and reading before bed. It can be hard to start, but just do 5 minutes and grow your learning habit. And it can be overwhelming at first — you might feel like you don’t know anything. Trust me, we all feel that way. In the new digital age, you need to be a sponge. Let me know what I missed in the comments!

How to get your learn on.

11

how-to-get-your-learn-on-121acfbd0bfc

2018-03-09

2018-03-09 23:48:17

https://medium.com/s/story/how-to-get-your-learn-on-121acfbd0bfc

false

786

null

Machine Learning

machine-learning

Machine Learning

51,320

Dave Stepp

Director of Emerging Technology and Google Cloud lead at Accenture Federal.

d904b25af014

dstepp2

8

36

20,181,104

null

0

null

0

null

2017-10-27

2017-10-27 05:42:07

2017-10-24

2017-10-24 17:09:43

5

false

en

2017-10-27

2017-10-27 05:54:16

1

121bf624c3f8

3.818239

2

0

Move now to measure and manage the entire digital journey. This is the time to revert from optimisers back to creative thinkers.

5

Google Masterclass 2017: Our Key Insights Move now to measure and manage the entire digital journey. This is the time to revert from optimisers back to creative thinkers. Recently, we took to the Google Masterclass in Melbourne to learn more about marketing in an AI-first world. The LeadTech team gained insights into changes in optimisation, automation, measurement, and attribution. How can an agency of the future prepare for these changes? Here are our key takeaways. Caroline Campaign Manager At this year’s Google Masterclass, there was an emphasis on demonstrating how Google products are evolving to be smarter (and more agile). As a campaign manager, I’m also developing similar skills when managing my campaigns. The class was really helpful to explore how much time we can save through automating the majority of our tasks, and using a smarter setup when planning. Specifically, I really enjoyed learning more about customising ads for the audience who will see them. That means adjusting messaging, timing and placement to change whether the user is commuting on a Monday morning, or relaxing on the couch on a Friday night. While this may be more work in the setup phase, it leads to smarter bidding and a better customer experience. Brendon Director of Advertising This year’s Google Masterclass really eclipsed previous years in its content and delivery. Automation, Optimisation, and Machine Learning/Deep Learning were key themes with a strong emphasis on how to approach the platform’s new features to scale operations for agencies. In today’s hyper-competitive marketplace, the necessity to bring relevance and meaning to the consumer places more emphasis on evaluating how your advertising addresses their needs. We need to be dynamic, agile, and creative with our ad creation while using analytics (and machine learning) to pivot effectively. David Booth from Cardinal Path was particularly insightful on what features are available today to accelerate growth into the future. It was fantastic to attend an event that was less about the hard-sell, and more about embracing new features methodically and product adoption from all levels. My biggest takeaway: embrace change and the implementation of automation to achieve a scale that services more customers better. Stephania Campaign Manager Aisling Finch, Marketing Director, explained how we’re moving from a mobile-first to an AI-first world. A revolution driven by AI is already shaping many parts of our lives but how will it change the operations of agencies? Martin Curtis, Head of Performance Agencies, elaborated on this topic which has puzzled me ever since I first started my career. Connecting the dots between increasing data points is getting more and more complex. At the same time, we see agencies moving away from managing defined aspects towards mastering the entire digital journey. AI is fundamental in this transformation to navigate and control this increasingly complex journey. The result? 50% of our time is cleared which can be re-invested into creative thinking and strategy rather than optimisation. Most agencies might not be there just yet, but future-proofing your agency for when the time comes is all about investing; invest in solid data management capabilities, invest in machine learning, invest in creative thinking & strategy. And remember the latter is everyone’s business in your agency. Marina Senior Campaign Manager Unless you’re a data scientist, attribution modelling is a pain — so the product I’m most excited about is Google Attribution. The aim of Google Attribution is to measure the “true” customer journey, which extends beyond digital. In the event, David Booth illustrated how the current attribution system is flawed in a fun way, by having a few members of the crowd pass a Chromecast up to the front, and letting the last person to touch it keep it. We’re making improvements in analytics and attribution a priority by exploring how our campaigns are affecting customer behaviour offline so that all our marketing efforts can be working holistically. It’s great to see Google reducing reliance on last click, so we can gain a deeper understanding of our customers, and deliver them a better experience. Where can you start? Creativity and strategy need to be everybody’s responsibility. Start by retaining the best talent, hiring great people, and developing your people; creative thinking and strategy skills. Google tells us to embrace change, be curious and live life as a student in order to progress. But how can you integrate this approach into your team? First up, build measurement, data and feed capabilities. Then, automate everything. This gives your coworkers more time to dedicate to creative solutions and strategy expertise. It’s also important to diversify your capabilities. Explore videos, feed management or even programmatic advertising. Above all, think broad and know what you want. Originally published at socialgarden.com.au on October 24, 2017.

Google Masterclass 2017: Our Key Insights

34

google-masterclass-2017-our-key-insights-121bf624c3f8

2017-10-30

2017-10-30 00:45:39

https://medium.com/s/story/google-masterclass-2017-our-key-insights-121bf624c3f8

false

791

null

Google

google

Google

35,754

Social Garden

We are a company that changes the way sales and marketing work together 🌱

f5ed7de0bb34

socialgarden

12

1

20,181,104

null

0

null

0

null

2018-08-12

2018-08-12 10:51:56

2018-08-12

2018-08-12 11:32:46

2

false

en

2018-08-12

2018-08-12 11:32:46

2

121d84616b4c

2.700314

3

0

Subjective metrics (e.g. happiness) alone aren’t very consistent or reliable data points. Objective metrics (e.g. distance run) are great…

5

On the Importance of Subjective and Objective Measures Subjective metrics (e.g. happiness) alone aren’t very consistent or reliable data points. Objective metrics (e.g. distance run) are great for statistical analysis but don’t yet exist for many of the things we care about. To get around that, we use markers or proxies of the things we care about. I argue that a combination of subjective and objective metrics is a powerful tool in gaining insights. Here’s an example: The latest version of my personal tracking framework has around 40 data points for both Subjective Productivity Rating (SPR) and for Deep Work hours. SPR is a measure of how productive my day was with reference to the things most important to my life and career, on an integer scale of 1–5. Deep Work hours are a related objective metric, simply measured by how many hours I logged in my retrospective calendar as being spent in a state of Deep Work. This is really only a proxy for how productive I’ve been, as my hours in this focused state are not always equally intense, so an hour of Deep Work is of varying impact. On their own, neither is a very good indicator of how productive I’ve really been. SPR is subject to bias, flawed perceptions and the inherent noisiness of quantifying a psychological experience, whilst Deep Work hours are only a reference point for how productive I was. Both are great things to track individually, but together they allow me to get a sense of: How productive I perceive my time spent working to be. If/when more time in deep work starts having diminishing returns. How closely each may be tethered to my actual productivity. Looking at the data SPR vs Deep Work (h) for 40 days As you can see (from the limited data I’ve gathered in this new framework), hours spent in Deep Work on any given day (X-axis) and SPR for a given day (Y-axis) only correlate with R² ≈ 0.41 on a linear fit. This isn’t very high, but visual inspection of the graph reveals more insights. Days of fewer than 2 hours in deep work have SPR values ranging from 2–5 in a clustering pattern. Then there is a huge gap in the data from 2 to 4 hours. I clearly get into a flow state after 2 hours of work and then just keep going for 4+ hours. On the upper end, when I’ve spent 4+ hours in a Deep Work state, my subjective experience is always one of high productivity – most values are 5/5. Takeaways Spending more time in a state of Deep Work makes me perceive myself as having been more productive. It also makes my perceived productivity more consistent. This could be due to the fact that I’m actually more productive, or merely a subjective experience of being more productivity. I tend to work deeply for less than 2 hours or more than 4 hours, but seldom in between. This is likely due to entering the flow state. The expected positive relationship between SPR and Deep Work hours is seen and of reasonably-high correlation given the noise levels of subjective data measures. Ideally, I need some concrete measure of my productivity – e.g. lines of code committed, words published on Medium, assignments completed, etc. to compare with both these measures to validate how accurate my perception is and how effective my deep work truly is. This is something I’ll probably look into when I’ve got more data. Self-tracking is as much of an art as a science at this point. I hope some of the ideas presented here inspire you to track and quantify yourselves in new and helpful ways.

On the Importance of Subjective and Objective Measures

7

on-the-importance-of-subjective-and-objective-measures-121d84616b4c

2018-08-12

2018-08-12 11:32:47

https://medium.com/s/story/on-the-importance-of-subjective-and-objective-measures-121d84616b4c

false

614

null

Productivity

productivity

Productivity

96,541

Quantified Vagabond

Biohacker, computer scientist, and self-experimenter

3f976d1402aa

quantified.vagabond

4

8

20,181,104

null

0

null

0

null

2018-07-24

2018-07-24 11:55:21

2018-07-24

2018-07-24 12:29:00

0

false

en

2018-10-14

2018-10-14 05:13:27

3

121daf121078

1.090566

1

0

This day’s Machine learning is so pervasive that you probably use it dozens of times a day without knowing it. Many researchers also think…

1

The 7 Secrets You Will Never Know About Machine Learning. This day’s Machine learning is so pervasive that you probably use it dozens of times a day without knowing it. Many researchers also think it is the best way to make progress towards human-level artificial intelligence. its helps to push relevant advertisements based on users search history by Google and Facebook. Machine learning is used to handle multi-variety data in dynamic environments. its do Fast Processing and Real-Time Predictions and also offers a continuous quality with big process environments. the process of automation of tasks is easily possible by machine learning. and its help to enables computer systems to learn, improve, and essentially “evolve” as they are exposed to new data. The power of Machine Learning is hidden in the self-teaching algorithms, which when exposed to data, can study and learn for improved best results. Machine learning and cognitive computing also empower the chat box to know what it can and to pass interactions on to human agents when it has a low level of confidence in providing the correct solution. They can lower wait times and also increase customer satisfaction. Madrid software training Solutions is one of the best machine learning institutes in Delhi that offers Machine Learning Courses in Delhi. Machine learning is the part of artificial intelligence that helps to implement the training data with the right combination of algorithm tool or application that can work while learning the behavior of the users and respond accordingly. The machine-learning is playing a significant role in expanding the growth of the companies and optimizing their work process, so in order to find how Machine Learning helping companies to improve the work processes also.

The 7 Secrets You Will Never Know About Machine Learning.

1

the-7-secrets-you-will-never-know-about-machine-learning-121daf121078

2018-10-14

2018-10-14 05:13:27

https://medium.com/s/story/the-7-secrets-you-will-never-know-about-machine-learning-121daf121078

false

289

null

Machine Learning

machine-learning

Machine Learning

51,320

Sunil Upreti

null

3cc663dfff14

sunilseo30

14

63

20,181,104

null

0

null

0

null

2018-04-03

2018-04-03 13:52:45

2018-04-03

2018-04-03 14:02:36

1

false

en

2018-04-03

2018-04-03 14:02:36

3

121e23857e61

5.177358

1

0

72% of Singaporeans are “very or somewhat concerned about being able to live comfortably in retirement” and target an annual return of 8.4%…

5

Augmented Intelligence will make Financial Planning accessible to all Photo by Olu Eletu on Unsplash 72% of Singaporeans are “very or somewhat concerned about being able to live comfortably in retirement” and target an annual return of 8.4% across all their savings and investment products according to a survey from Blackrock in 2016. But they hold about 48% of their savings in cash. In France, 70% of the respondents surveyed by Deloitte think that the current retirement funding system based on inter-generational solidarity will not last and that their pension will only cover 2/3 of their needs (2016). In the USA, the same year, a GoBankingRates report reveals that 56% of Americans have no- or less than $10,000 retirement savings. Different countries, different policies, different cultures. Same aspirations. But same conclusion: when it comes to retirement, governments fail to address their citizens’ needs. Therefore, many are left on their own when planning for their future, not to mention various additional financial uncertainties: purchasing a house, children’s education costs, health issues, accidents, unemployment… The private sector has its own flaws: opaque, complicated and often biased. Financial Advisors are mostly commission-based, “free” to the consumer, but in fact acting as the sales arm of financial institutions which are incentivized to push their own financial and insurance products that are not always aligned to the customers’ goals and needs. Fee-based advisors’ services remove this bias but few people are willing to pay for independent advisory. What it tells us is that: Investing is a necessity for all, Most people do not have a plan to address a growing financial gap, There is a lack of financial literacy but also an opacity that benefits the sector, There is a necessary human element in the advisory process. Since Artificial Intelligence in the true sense does not exist yet, a better choice of word at this stage is Augmented Intelligence: technology, data and algorithms can be used to enhance our abilities as humans to comprehend, interact with and decide upon the world around us. So how can Augmented Intelligence make financial planning more accessible to all? Robo-advisory consists of the automation of the financial planning process: customer profiling, goal setting, risk assessment, insurance policies recommendations, portfolio optimization and periodic rebalancing based on market conditions to stay aligned with the financial goals. Automating this process will need to rely on more than simple hard-coded rules to be relevant and personalized. It will have to continuously learn to match customers goals and risk aversion to protection options and investment portfolios in the customer context; it will be able to “nudge” the customer to trigger action, as well as anticipate and respond to market events. In short: robos bring intelligent automation to the advisory process. According to Statistica, the average assets under management (AUM) per user managed by “Robo-Advisors” in 2018 amounts to US$40,420 in the USA, $15,803 in France and $14,379 in Singapore. To put things in perspective, by 2025 the total market size of hybrid robo services will increase to USD 16,300 billion, which constitutes just over 10% of the total investable wealth. Not all robo-advisory companies are made equal though: depending on their license, their operating model may go from a technology enabler (white labelled to financial institutions), to distributors (selling products from other financial institutions), to fund manager (ability to create and sell their own funds). Hence, some companies target consumers directly and attempt to break into the market with a super-low cost, fully automated offering, while others act as intelligent support systems to Financial Advisors and Wealth Managers -giving them evaluation and mapping tools to manage more customers better. Customer profiling and Portfolio Management are two areas of financial planning that can be significantly improved with Augmented Intelligence. Today’s customer profiling and risk assessment is ridiculously lengthy, unsophisticated and inefficient and barely protects the novice investor from him/herself: a typical client questionnaire includes more than fifty questions and relies on the judgement of a human advisor to assess the customer risk appetite or aversion in order to provide the portfolio that best suits him/her. As inventor Henry Ford put it: “if I had asked my customers what they wanted, I would have built a faster horse.” Similarly asking someone if he/she is ready to lose 5% or 90% of his/her investment value may not be the best way to get the full extent of the risk profile of an individual. But here is what can be done. First, build quantitative behavioral economic models using decision games that reveal true preferences and calculate risk tolerance and loss aversion scores. Second, identify patterns in financial plans from people with similar socio-economic background so they can benchmark theirs against their own peer group. Third, make this process better over time with machine learning by correlating the decisions made by each customer along the advisory journey. These intelligent and personalized nudges along the investment journey can make a tremendous difference in building trust and credibility while creating a sense of urgency and action. Portfolio management can also be enhanced and optimized: assets allocation is based on client risk profile and interests: geography, currency, industry or even thematic such as environmental, social and corporate governance — ESG, China’s one-belt-one-road, autonomous vehicles, etc. Whereas passive investors favor “lazy man” portfolios made of simple three or four tier index funds, active investors want more granular control over their investments. Such portfolios composed of diverse securities need to be tracked against market events which represents tremendous challenges today. As Blackrock puts it: “in a world that now sees an average of 4,000 brokerage reports a day comprising 36,000 pages in 53 languages, advanced text analysis is a necessity:” In particular, neural networks used in natural language processing provide new opportunities to augment traditional quantitative financial analysis with qualitative insights derived from the ingestion and processing of market news at scale such as sentiment analysis, institutional investors decisions patterns and market trends. This creates unique differentiation opportunities for both incumbents and new businesses by providing hedge fund grade tools directly to end-consumers -or to financial advisors and wealth managers to help maximize profits for their customers. It also opens a new door for advisory services pricing models based on true investment performance as opposed as commission based, asset under management based or fee based models. Moving to hybrid or fully automated advisory services with augmented intelligence requires removing some barriers and revisit regulations and licensing schemes: Incentives and commissions should be revisited to encourage performance-based pricing and direct sale of products in a more fair, transparent and open manner. The duty of care should apply to robos and algorithms which requires what some calls an “Explained AI”: automated tasks requires little explanation but automated reasoning requires explanations to be trustworthy, traceable and auditable. Investors will have to overcome a psychological barrier to invest their life long savings in some obscure start-ups, or may favor instead sometimes less sophisticated but more reputable institutions. In short, technology should not be replacing advisory human capital but rather augment its intelligence, abilities and productivity to serve more customers better, making financial planning more accessible to all. Damien KOPP, Head of Products BlueFire AI & Live Withe AI Board Member About Live With AI: Live with AI is a non-profit foundation based in Singapore. The foundation gathers thought leaders, decision-makers and French, Singaporean, and international researchers to lead working groups and research projects on the positive impacts of artificial intelligence to our society. The Live with AI community takes advantage of a presence at the heart of the South-East Asia region and an access to several research laboratories to issue recommendations which can be immediately applied and tested among very diverse communities looking for technology disruption. Live with AI is an independent initiative created at the occasion of the France Singapore year of Innovation 2018.

Augmented Intelligence will make Financial Planning accessible to all

1

augmented-intelligence-will-make-financial-planning-accessible-to-all-121e23857e61

2018-04-05

2018-04-05 10:09:17

https://medium.com/s/story/augmented-intelligence-will-make-financial-planning-accessible-to-all-121e23857e61

false

1,319

null

Investing

investing

Investing

51,660

Live With AI

A program aiming at understanding humans’ need towards the rise of AI and supporting our society to anticipate changes to better Live With AI.

ec3dcdf4dd73

livewithai

6

1

20,181,104

null

0

null

0

null

2018-02-13

2018-02-13 16:25:23

2018-02-13

2018-02-13 16:37:18

0

false

en

2018-02-13

2018-02-13 17:03:48

0

121ea9d3b392

3.50566

0

If you go and take a relaxed, well paid, 9–5 early in your career, you are getting paid to forgo the growth in your capabilities. Ok…

4

How to pick your first job or company: intelligence compounds If you go and take a relaxed, well paid, 9–5 early in your career, you are getting paid to forgo the growth in your capabilities. Ok, that’s fine, many people take that deal. More so in Germany, the number one country in the risk aversion scale (well, it would be off the scale if there was one). You may have better things to do with your time than just getting smarter. But then, if intelligence compounds (as in ‘compound interest’)… Your choice has a dramatic effect mid or long term. Your employer should pay you a lot more for that opportunity that you are forgoing. If you take a job at say Deutsche Bank (I could not think of a shittier company) and stay say seven years you are virtually accepting you are never going to be competitive, even if you may have been at the beginning of these seven years. Someone who used these seven years in situations with high growth opportunities will be very well positioned after that to solve really significant problems. Because intelligence compounds, it’s essential to pick the right starting job in your career. Or the right startup to found if you are planning your life as an entrepreneur. Because we spend so much of our time at work, one of the most powerful leverage points for increasing our learning rate is our choice of a work environment. How can we translate this observation into tactics? Optimize for growth Everyone who is any good at anything optimized for growth early in their career. Sometimes without knowing. Small, but constant, daily growth leads to dramatic changes long term. The good news is that you don’t have to think about the long-term plan. Just make sure that today was not wasted, and that you are learning at say a 1% rate per week. That is enough. It’s vital that you measure your improvement. Track the time it takes you to finish simple programming tasks. Then plot progress every year or so. Measuring progress doesn’t work well for creative things like designing architecture or debugging some difficult bug. But it’s worth trying. You can also ask others for honest feedback. Am I getting better at this? Often people won’t tell you to your face when you are getting ‘kinda good’, but they talk to each other. It’s only by chance that you realize you are growing a reputation. Culture A work environment that iterates quickly provides a faster feedback cycle and enables you to learn at a quicker rate. Lengthy release cycles, formalized product approvals, and indecisive leadership slow down iteration speed; automation tools, lightweight approval processes, and a willingness to experiment accelerate progress. Anything that works at ‘Deutsche bank speed’ is to be avoided. Do people help each other to get better? Or is the culture one of ‘one-up-manship’? Do the guys that have mastery share what they know? Are there others in your team with complementary skillsets? People Empathy is a skill like any other; it improves with practice. Reading fiction improves empathy. But the best book I can think of is Marshall Rosenberg’s Nonviolent Communication (NVC). If you can speak NVC you will be a fantastic manager; probably a tremendous father, partner, anything. Bradford Cross recommends it, and I can see why. I’ve read it, and I’m extremely impressed by Rosenberg’s ingenuity. It does take practice to get it to work. Another technocrat that recommends improving empathy is Chad Fowler. Bosses: make sure you have a good one, and that you understand what each other needs. People leave bosses, not companies. If you have a bad feeling during interviews, just say no. This single person can have an outsized effect on your growth and happiness. Toxicity Do good ideas get killed in front of everyone by management? Do bad ideas get the green light by sheer politics? Are you treated with respect when you deliver results? Do people admire that you are effective? Is incompetence tolerated? If so, run away. Having the freedom to fuck up and not been crucified is fantastic, but there are different types of fuck ups. Trying innovative things is bound to produce plenty of failures. But executing at world class level and failing is one thing. Failing to perform because of sheer incompetence: No. Everyone in the company should be able to tell the difference. The problem with tolerating incompetence is that it deters good people from doing their best. In fact, very likely they will leave. So finding the incompetent and ousting him or her should be everyone’s priority. One exception to this rule: polymaths, or well-balanced generalists. They are not world-class at any one single skill. A specialist may consider their competence not up to scratch. Generalists abound in early-stage companies. In fact, without them, there would be no startups, as everyone tends to wear many hats. The guy who can code, sell, and do strategy is worth his weight in gold even if you wince when you see his commit messages. Eventually, generalists get replaced as the company grows. This is painful for the generalist, so be gentle. People who can think strategically (e.g., find the intersection of which products can be built with current tech, and what the market is willing to pay for) are enormously valuable. More so in areas like machine learning, where everyone is trying to figure out what is plausible. If your company culture seems to be toxic for these profiles, it’s time to run.

How to pick your first job or company: intelligence compounds

0

how-to-pick-your-first-job-or-company-intelligence-compounds-121ea9d3b392

2018-06-08

2018-06-08 21:15:33

https://medium.com/s/story/how-to-pick-your-first-job-or-company-intelligence-compounds-121ea9d3b392

false

929

null

Careers

careers

Careers

35,661

Totem Code

Everyone is an autodidact in tech. But we can kick up your learning speed. Really learn deep learning.Emphasis on good engineering practices and production code

4497e8f8153c

deeplearningret

1

18

20,181,104

null

0

null

0

null

2018-08-31

2018-08-31 13:02:53

2018-08-31

2018-08-31 13:16:47

0

false

en

2018-08-31

2018-08-31 13:16:47

3

12200e4794fc

2.528302

1

0

Statistics is the study in mathematics and this explains about collecting, analyzing, interpreting, presenting and organizing of data…

5

How Statistics Gets its Right Introduction at the School Level? Statistics is the study in mathematics and this explains about collecting, analyzing, interpreting, presenting and organizing of data. Planning along with survey as well as experiment is very essential for students to know about a set of data to design and to represent. Statistics is a vast field in study and to understand it in a better way, students get introduction of statistics at the school level. For the right knowledge students must know about all basic terms and factors. It is important to know that statistical data means some set of numbers that need to calculate to obtain various values. What are the basic concepts in statistics? Statistics and its interpretation along with analyzing and representation depends on these following basic factors as – Arithmetic mean – Arithmetic mean is also known as simply mean and for a data set it is evaluated by dividing the addition of numbers in the set by the number of quantity. There are three different methods to calculate mean – Ø Direct method Ø Assumed mean method Ø Step Deviation method Ø Mean for grouped frequency data Median – Medicine is simply the middle value of a distribution set of data. Middle value means an exact amount of the distribution in two halves or equal parts. It means the number of observations above the median is same as the number of observations below the median. There are different factors on which median depends for continuous frequency distribution as – Ø Lower limit of the median class Ø Frequency Ø Cumulative frequency of the class preceding the median class Ø Size of the median class Ø Number of observations Mode – This is also known as the modal value of a distribution and it is represented as the value of the median in which the frequency value is maximum frequency. Anyone can find out mode for continuous frequency distribution through its suitable formula. Cumulative frequency distribution or graphical frequency distribution This depends on two different ways and these are – Ø Cumulative frequency distribution or graphical frequency ogive curve less than type Ø Cumulative frequency distribution or graphical frequency Ogive curve more than type Median of a ground data needs to be evaluated with graphical representation as x — coordinate of the point of intersection of the two different types as less than type and more than type. You can easily understand its finding way after evaluation and the exact formula is 3 Median = Mode + 2 mean How a student gets an exact way to understand all problems related to this subject? Different problems in statistics can easily be solved only when a student understands all basic concepts of it. Now, there are different formulas and each one must go through and take care of these formulas. Now, they must apply these formulas according to the requirement of each question. Students need to solve as much as they can do. Only after that they will get a perfect and suitable solution related to this topic, also they will be able to handle critical problems. Homework and assignments is provided to the students and these tasks are always beneficial to them. In case, they require assistance; then they can easily go with online homework help solution related to the topic without any hesitation. Requirement of statistics in different fields enhances its importance. In these days statistical computation makes it perfect and more convenient. The different subjects as economics, biology, geography, and others also get related with the different terms of statistics. So, various software programs are developed to perform various tasks conveniently. Now, you can easily get that a perfect knowledge from its introduction level will be there to give a good grip over the subject. At 24x7homeworkhelp.com, we strive to provide students with all-round academic support. Students can access our homework help service any hour of the day. 24x7 homework help always helps them to make submissions before the prescribed deadline.

How Statistics Gets its Right Introduction at the School Level?

1

how-statistics-gets-its-right-introduction-at-the-school-level-12200e4794fc

2018-08-31

2018-08-31 13:16:47

https://medium.com/s/story/how-statistics-gets-its-right-introduction-at-the-school-level-12200e4794fc

false

670

null

Data Science

data-science

Data Science

33,617

Evelyn W. Minnick

null

25936e3c3431

seoamit8129

0

1

20,181,104

null

0

null

0

null

2017-10-09

2017-10-09 02:31:47

2017-10-09

2017-10-09 02:57:31

4

false

en

2017-10-09

2017-10-09 02:57:31

4

12208002c1f1

3.118868

3

0

Satya Nadella’s “Hit Refresh” has been a great read. This is the first of two posts inspired by it.

3

Augmented Humanity Satya Nadella’s “Hit Refresh” has been a great read. This is the first of two posts inspired by it. I’ve been equal parts anxious and excited about the promise of AI. I knew that Satya Nadella would have a very balanced view, so after being thrilled by his description of quantum computing, I was eager to hear how he saw this new frontier playing out. One of the healthiest suggestions he has for a society grappling with the Rise of the Machines is that we get over our false “utopia/dystopia dichotomy.” It’s not either HAL or WALL-E. Most importantly, the place in the middle that we land is completely up to us. Instead of fretting over the future, Nadella entreats us, we need to spend more time inventing it. Augmented Humanity, first realized by Lee Majors and Lindsay Wagner In the midst of this discussion, he provides a description of the three core principles that are driving the AI efforts at Microsoft. The first resonated with me because I’d just come to the same conclusion hours before in an interchange with my Tweeps: Amy Cane Dolzine reminds us that authentic voices are still needed amid the synthetic ones. Ben (@ideocial) expresses some healthy cynicism about the ability to encode true empathy. As I read their exchange, for some reason I thought about how much I preferred the applications I’ve seen of Augmented Reality (AR) to those of Virtual Reality (VR), and then wondered if it wouldn’t help AI’s reputation to get a re-brand: Augmented Humanity. Hours later, I read in “Hit Refresh” about the first of Microsoft’s AI principles: “…we want to build intelligence that augments human abilities and experiences. Rather than thinking in terms of human vs. machine, we want to focus on how human gifts such as creativity, empathy, emotion, physicality, and insight can be mixed with powerful AI computation…” (p. 201) Yes! This is a great description of the true promise of AI. Not to usurp us, and not to lull us into an intellectual slumber, but to provide us with the extra “juice” to enhance those things that are truly, uniquely human. Satya goes on to talk about AI being the “third run time,” and my rebranding made even more sense. Technology has been mediating, and enhancing, our performance for decades. The first “run time” was the PC, and it made interacting with data much more effective and dynamic. The second “run time” was the web. This platform took our relationship with data to a new level, but it also enhanced our interaction with others. Of course, because of the impact of social media, whether it’s been an enhancement or a degradation is a subject of debate, and I’ll come back to that. The third “run time” is AI. It holds the promise of taking over much of the interaction we were having to do in the data arena, allowing us to focus on the further enhancements it can also make in our human interactions. I think the possibility of an Augmented Humanity through this next-level computing is exciting, but here comes a caution for those still struggling with that second run-time. If you haven’t figured out how to make the most of the “social” wave — how to use the tech now available to maximize the power of your inner- and outer-loops — then the multiplying power of AI will provide a much smaller return for you and your work. If you are using collaboration techniques that aren’t growing your human network, and aren’t encoding those relationships digitally, then AI’s neural network will have less “raw material” to work with in enhancing those relationships. Think about the tools you use not only in terms of how they will help you get today’s job done more efficiently, but about how they will multiply your ability to get tomorrow’s job done.

Augmented Humanity

12

augmented-humanity-12208002c1f1

2018-06-20

2018-06-20 03:19:32

https://medium.com/s/story/augmented-humanity-12208002c1f1

false

641

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Chris Slemp

Improving employee engagement with better communication, transparency, and responsiveness. Customer Success Manager at Microsoft UK.

7dd50697d65a

cslemp

385

316

20,181,104

null

0

FROM ubuntu:16.04 ... RUN pip --no-cache-dir install \ ipykernel \ jupyter \ numpy \ pandas \ sklearn \ && \ python -m ipykernel.kernelspec ... # Install TensorFlow CPU version from central repo RUN pip --no-cache-dir install \ http://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.0.0-cp27-none-linux_x86_64.whl ... CMD ["/run_jupyter.sh", "--allow-root"] docker pull tensorflow/tensorflow $ docker pull tensorflow/tensorflow Using default tag: latest latest: Pulling from tensorflow/tensorflow 3b37166ec614: Already exists ba077e1ddb3a: Already exists 34c83d2bc656: Already exists 84b69b6e4743: Already exists 0f72e97e1f61: Already exists 6086c6484ab2: Pull complete 25817b9e5842: Pull complete 5252e5633f1c: Pull complete 8de57ae4ad7d: Pull complete 4b7717108c3b: Pull complete b65e9e47e80a: Pull complete 006d31e013ea: Pull complete 700521cc53f3: Pull complete Digest: sha256:f45d87bd473bf9992414f8d736a808277b4f3e32159566 Status: Downloaded newer image for tensorflow/tensorflow:latest $ docker images tensorflow/tensorflow REPOSITORY TAG IMAGE ID CREATED SIZE tensorflow/tensorflow latest 76fb62c3cb89 2 weeks ago 1.23GB docker run -it -p 1234:8888 tensorflow/tensorflow docker run --help docker pull leemeng/cat docker run -it -p 2468:5000 leemeng/cat FROM python:3.6.3 MAINTAINER Meng Lee "b98705001@gmail.com" COPY ./requirements.txt /app/requirements.txt WORKDIR /app RUN pip install -r requirements.txt COPY . /app ENTRYPOINT [ "python3" ] CMD ["app.py"]

14

null

2018-09-08

2018-09-08 17:14:18

2018-09-10

2018-09-10 11:49:14

13

false

zh-Hant

2018-09-10

2018-09-10 11:49:14

32

122140cd6a68

4.739623

2

0

3 種活用 Docker 的方式（上）

3

給資料科學家的 Docker 指南 3 種活用 Docker 的方式（上）今天我們來聊聊如何將 Docker 應用在資料科學領域裡頭吧！全文共分上下 2 篇。在這篇裡頭，我們將透過一些簡單的比喻來直觀地理解何謂 Docker，並讓你能在閱讀本文後馬上利用 Docker 來加速你的開發效率；在下篇的內容當中，我則會分享一個資料科學家（Data Scientist：DS）為了解決一些數據問題而時常碰到的 3 種 Docker 使用方式。不管是哪一篇，我們都不會深入探討 Docker 本身是以什麼技術被實現的。反之，我們將會以 DS 的角度，專注在「應用」層面：如何把 Docker 實際應用在資料科學以及資料工程領域裡頭。這系列文章適合 2 種讀者：對 Docker 完全沒有概念但想讓自己的 Workflow 更有效率的資料科學家熟悉 Docker，但好奇其在資料科學領域如何被應用的工程師讓我們開始吧！雲端運算 & Docker 在解釋何謂 Docker 之前，讓我把你已經非常熟悉的雲端運算（Cloud Computing）老朋友叫出來。 Amazon Web Service（AWS）、Google 雲端平台（GCP）以及 Microsoft Azure 大概是大家最耳熟能詳的幾家雲端計算 / 服務平台了。隨著時代的演進，這些平台提供越來越多樣的機器學習 API，讓開發人員不需做複雜的開發，透過一個 HTTP 要求就能直接使用各種酷炫的服務，比方說： Amazon Lex 讓你使用 Amazon Alexa 的深度學習技術建立聊天機器人 Google Cloud Vision API 讓你快速建立一個圖像辨識服務 Azure Content Moderate API 讓你自動審核網路上的圖片以及文字儘管如此，很多時候只使用這些現成的 API 並不能滿足我們這些 DS 以及企業的野心。比起使用現成 API，如何運用雲端運算來 scale 各種數據處理工作是一個 DS / DE 更常問的問題除了直接用各家雲端平台提供的 API 以外，一個 DS 可能更常需要利用雲端上的計算資源來完成以下的工作：部署一些新的分析工具來嘗試提升自己及分析團隊的效率開發、訓練、部署並規模化（scale）自己的機器學習模型對大量數據做批次處理，將結果儲存後顯示在儀表板上事實上，這就是本系列文章最想要跟你分享的 3 件 DS 可以活用 Docker 來最大化產出的案例。當我們透過這篇文章（上篇）熟悉了 Docker 的基本概念以及操作以後，就能在下篇裡頭深入地探討它們。因此在這篇先讓我們專注在學習 Docker 的基礎知識吧！雖然我們現在不會細談，但如果你再看一次上面的 3 個工作的話，會發現裡頭可不只包含資料科學（Data Science）。除了建置儀表板以及設計 ML 演算法以外，這裡頭還包含了不少軟體工程、資料工程甚至 DevOps 成分。當然資料工程師（Data Engineer）很樂意幫助你，但如果你想要快速地自己兜出一些方法呢？你該用什麼工具？你可能覺得一個 DS 要在各種 deadlines 內完成以上所有的事情是不可能的。不過後面我們會慢慢發現，活用 Docker 能讓這些工作變得簡單許多。接著就讓我們以 DS 的角度了解 Docker 到底是什麼技術。我相信閱讀接下來的文章，對你之後開發效率的提升是一個非常好的投資。 Docker：可愛的大鯨魚首先看看以下這張 Docker 示意圖：有什麼感覺嗎？注意到上圖包含了 3 個要素：海洋鯨魚貨櫃現在讓我們發揮點想像力。如果你把雲端運算的平台想像成一個充滿運算資源的大海的話，Docker 就是如圖中在裡頭悠遊的大鯨魚。這隻鯨魚將上述所有 DS 想要做的數據處理工作、執行的 App，一個個封裝成彼此獨立的貨櫃，並載著它們在這大海上運行。 Docker 提供的抽象化讓我們能輕鬆地運行任何想使用的資料科學工具、軟體而不需花費過多時間在建置底層環境。我知道你可能還是沒什麼感覺，讓我們看下去。鯨魚背上的貨櫃：Docker 容器實際上這一個個假想的貨櫃就代表著 Docker 術語裡頭的容器（Container）。「容器」顧名思義，是一個「容納」了某些東西的「器具」。一般而言，一個容器裡通常會包含了一個完整的 App。這邊的 App 不是手機上的 App，而是指廣義的應用程式（Application）。 DS 常用的 App 可以是：一個包含 TensorFlow 函式庫的 Jupyter Notebook 伺服器一個 ML 產品，如透過模型來判斷圖片裡頭有沒有貓咪的 Flask App 一個 SQL 查詢以及資料視覺化的工具，如 Superset 一個簡單的 Python Script，針對輸入的大量數據做處理要從頭建構這些 App 的環境不是不可能，但除了基本的 pip install 以外你還需要花不少工夫；更令人困擾的是，很多時候你在 Mac、Windows 上安裝環境的步驟，到了雲端上的 Linux 機器上就完全行不通了。如果這時候有人先幫我們把一個在哪邊都能跑的 App 環境建好，我們不是就能馬上開始使用各種分析工具，進行各種有趣的分析，而不用煩惱底層如不同 OS 的差異了嗎？ Docker 的容器就是這樣的一個概念，幫你事先將一個 App 所需要的所有環境，包含作業系統都「容納」在一起。 Docker 將一個 App 會使用到的程式語言函式庫（JAVA、Python、R）、資料庫、甚至作業系統（OS）都包在一個自給自足的容器（CONTAINER）裡頭。想使用某個 App 的 DS 不用從頭建置環境，只需利用 Docker 啟動該容器即可開始工作容器裡頭不只包含 App 自己本身的程式碼，也涵蓋了所有能讓這個 App 順利執行的必要環境： App 需要的各種 Python 函式庫，如特定版本的 TensorFlow、Pandas 及 Jupyter Notebook MySQL、MongoDB 等 App 會用到的資料庫 App 會用到的各種 metadata、資料集各種 OS 限定的驅動程式（drivers）、依賴函式庫（把所有你想得到的東西填進來）包羅萬象。因此只要我們能利用 Docker 把一個 App 需要執行的環境全部包在一個容器裡頭，我們就能在任何有 Docker 的地方啟動並運行該容器。不再需要每次重新建置環境，也不用考慮不同機器上的安裝問題。而這正是 Docker 最強大的地方： Docker — Build, Ship, and Run Any App, Anywhere 因為連 OS 都被包起來了，實際上每個容器（container）的執行環境都是自給自足的（self-contained）。你可以把它想像成非常輕量的虛擬機器，其執行結果不會因為啟動該容器的「計算環境」不同而受到影響，在任何地方（Anywhere）都能順利被執行，且執行的結果都是一樣的。以我們前面的比喻來說的話，每個貨櫃（容器 / App）都是我們想要 Docker 幫我們運送（執行）的東西，而不管 Docker 這隻鯨魚（或大船）現在在哪個海洋（計算環境）裡頭，它都能使命必達。 Docker 就像艘大船，幫我們在任何海洋（計算環境）上運送我們的貨櫃（容器）有一點值得澄清的是，就算 Docker 幫我們抽象化建置一個 App 環境的工作，在執行一個容器的時候，我們還是需要實際的計算資源來跑這些容器。因此前面所謂的「計算環境」指的是一個擁有計算資源（CPU、GPU、記憶體 etc）且我們實際運行 Docker 的地方。這計算環境可以是任何一家雲端服務平台上的機器，如 AWS 的某台 EC2 機器、GCP 上一個包含數千台機器的群集（Cluster），或是你現在用來看本文的筆電。只要 Docker 能在該計算環境下運行，它就能幫我們在該環境「之上」執行任何容器。簡單來說： Docker 幫我們抽象化在任何 OS 上建置環境的工作。只要給 Docker 一個容器，它就能在任何地方啟動該容器以供你使用。現在你對 Docker 以及容器概念有個高層次的理解了，讓我們來看看這些 Docker 容器實際上是怎麼來的吧！貨櫃（Docker 容器）從哪來在了解 Docker 這隻大鯨魚能幫我們運行任意的容器 / App 以後，你腦中浮現的第一個問題應該是：這些容器（貨櫃）最初是怎麼被產生的？非常好的一個問題。事實上，要產生一個新的 Docker 容器，Docker 需要一份「環境安裝步驟書」來讓它幫我們自動地建置容器內的環境，比方說使用什麼 OS，用什麼版本的 TensorFlow 等等。這份步驟書在 Docker 的世界裡被稱作 Dockerfile。舉個例子，以下是 Tensorflow 官方釋出的一個 Dockerfile（截錄重要部分）：除了 RUN、CMD 等 Docker 專屬的關鍵字以後，你會發現這份 Dockfile 裡頭的指令其實跟你平常在本地開發時也會使用的指令如 pip install 沒有相差太多。差別在於透過第一行的 FROM ubuntu:16.04 指令，我們要求 Docker 在這個容器裡頭建置一個 Ubuntu OS 後，在其之上安裝這些函式庫。追求規模性：Docker 映像檔的誕生聽完以上的解釋，你可能會覺得在我們每次要啟動一個新的容器的時候，Docker 就得拿出 Dockerfile，一步步建置該容器的環境。這樣的實作也不是不行，但很沒有效率。為什麼？其中一個考量是可擴展性（Scalability）。有時你會想要用同一份 Dockerfile 在短時間內迅速地產生好幾個一模一樣的容器(s)：用多個相同的機器學習模型，同時對大量的新數據做批次預測使用多個相同的 Python Script 來處理大量數據這時候與其在每次要啟動新的容器時才拿出 Dockerfile 建置環境，Docker 可以事先用這個 Dockerfile 把建置環境所需的步驟先做好一遍，然後把該環境「拍張照」，存成一個 Docker 映像檔（image）後等待之後的使用。等你決定要開始使用容器的時候，因為我們已經有一個環境的快照（Snapshot），Docker 就能利用該映像檔，快速地啟動 1 個（或 100 個）相同的容器給你。 Docker 三元素： Dockerfile、Docker 映像檔以及 Docker 容器到了這邊，我們已經了解 Docker 最基本也是最重要的概念： Docker 利用 Dockerfile 預先建置好一個 Docker 映像檔。在使用者想要使用容器的時候，以該映像檔為基礎，運行一個對應的 Docker 容器坐而言不如起而行。在掌握了這些概念以後，我相信你也迫不及待地想要開始使用 Docker 了，接下來就讓我們實際操作 Docker 來體會一下它的威力。 Docker 映像檔：法式千層酥不管是 Windows 或是 Mac 用戶，你都可以很輕鬆地在官方網站下載 Docker 並安裝。下載完以後啟動 Docker，大鯨魚就會在你的筆電上開始閒晃，等待你的指示。一般而言，我們會在 terminal 使用各種 docker 指令來跟大鯨魚溝通。當 Docker 就緒以後，依照我們前面的所學，你會需要一個 Dockerfile 或是 Docker 映像檔來產生一個 Docker 容器。就像 Github 是一個被大家拿來分享程式碼的地方，Dockerhub 則被用來分享 Dockerfile 以及 Docker 映像檔。假設我們現在要開始一個新的 TensorFlow 專案，並且想透過 Jupyter Notebook 進行開發，最省力的方式就是從 Dockerhub 下載一個 TensorFlow 官方幫我們弄好的 Docker 映像檔。讓我們打開一個 terminal 並輸入 docker pull 指令：第 1 個 tensorflow 代表 Tensorflow 的官方 Dockerhub repository，就跟 Github repository 的概念相同；第 2 個則是容器名稱。你會看到當 Docker 在下載映像檔的時候，同時也在建置環境，而其環境會分成一層一層（Layer）的：我們不會細談 Docker 實作細節，但你可以想像 Docker 映像檔是一個法式千層酥（Mille Feuille）。這時候的 Docker 是一名蛋糕師傅，利用 Dockerfile 作為食譜，逐行執行裡頭的指令以建立一層層的環境。每做出一層新的環境，就把它加在目前所有環境的上面，最後成為一個 Docker 映像檔。這樣做有 2 個好處：當你對 Dockerfile 做變動的時候，Docker 可以只針對被改變的那一層環境做修改，而不用重建每一層，減少建置環境所需要的時間有利用到一樣環境的不同映像檔可以分享部分結果（如上面的 Already exists）一個 Docker 映像檔就像是蛋糕師傅利用 Dockerfile 食譜做出來的法式千層酥（誠摯地希望你不是晚上看本文，餓了）依照你的網路速度，下載映像檔所需的時間可能有所不同。在下載完成以後，輸入 docker images 指令可以顯示所有目前本地端擁有的 Docker 映像檔：這邊因為我的環境裡已經有一大堆的映像檔，我在 docker images 後面加入額外的篩選器來告訴 Docker 只顯示 tensorflow repository 裡頭的 tensorflow 容器。有了映像檔以後，最令人期待的時刻終於來臨了！我們現在要呼叫 Docker 幫我們從這個映像檔產生並執行（run）一個新的 Docker 容器：短短一行指令，包含了 3 個你不可不知的重要概念：利用 docker run 來告訴 Docker 我們要利用 tensorflow/tensorflow 映像檔來運行一個容器。實際上 Docker 容器就是在 Docker 映像檔的環境之上再加 1 層可執行的環境供你使用（貫徹千層酥的理念）利用 -it 參數來告訴 Docker 我們同時要建立一個互動式的 TTY 連線，讓容器內的結果直接顯示在我們的 terminal 裡頭，彷彿我們在本地環境下執行該 App 一樣。我們之後還可以直接在 terminal 使用 Ctrl + C 或 Command + C 來終止容器利用 -p 1234:8888 告訴 Docker 我們將會透過本地端的 1234 port 來連到容器裡頭的 8888 port 你可以透過來查看所有 docker run 可以使用的參數。另外，一個 DS 應該都知道，8888 是 Jupyter Notebook 預設的 port。因此我們的企圖就跟司馬昭之心一樣，打算透過本地端的 1234 port 連到在容器裡頭跑的 Jupyter Notebook。現在打開你的瀏覽器並輸入 localhost:1234，應該就能連到容器內部的 Jupyter Notebook 伺服器：容器內的 Juypter Notebook 畫面，所有環境包含 TensorFlow 都已經幫你設置好（輸入你在啟動容器的 terminal 裡看到的 token 就能通過認證）對你沒看錯，你已經用 Docker 建置了一個完整的資料科學環境，裡頭有 TensorFlow 以及 Jupyter Notebook。而你只需要 2 個指令： docker pull tensorflow/tensorflow docker run -it -p 1234:8888 tensorflow/tensorflow 建置環境什麼的交給 Docker 吧，你已經能馬上開始實作機器學習模型了。有些 DS 可能會覺得他的 Anaconda 或者是 pip 功能爐火純青，不需要用到 Docker 也能自己在本地建出這樣的環境。其實沒錯，如果你只是開發個人專案，說真的不學 Docker 也沒關係（喂！）但就如我們在下篇會看到的，當你在開發企業等級的數據處理工作、機器學習模型的時候，你可不能永遠躲在你的本地環境裡頭。當你習慣於在不透過 Docker 的情況下在本機建置環境，等到要在各種雲端平台上的機器重現你的結果的時候，你就會發現不妙了。利用 Docker 分享你的成果為了加強你使用 Docker 的動機，讓我再給個例子。有持續關注我文章的讀者會發現，我在資料科學文摘 Vol.3 Pandas、Docker 以及數據時代的反思裡頭有提到，Docker 除了讓我們免除建置環境的痛苦以外，也能讓我們與他人簡單地分享開發結果。 Cat Recognizer 是我用 TensorFlow 以及 Flask 實作的一個非常 naive 的貓咪辨識 App。如同我們前面所說的，我事先將所有此 App 需要的環境用一個 Dockerfile 定義、全部包在一個 Docker 映像檔後分享在 Docker Hub 上面。任何想要使用此 App 的人，只需要利用 Docker 輸入兩行指令：接著他們就能在瀏覽器輸入 localhost:2468 來看到我的 App： Docker 讓你與其他人分享成果，不須額外做一大堆環境設定當然這個 ML App 在預測能力以及 UI 上都不完美，但這邊重點是你能利用 Docker 與他人快速地分享成果。如果你有想到其他利用 Docker 封裝好的 ML App 例子（或者是你接下來打算做一個自己的），非常歡迎留言讓我知道它們的存在：）總結呼！看完本文以後，相信你現在應該對 Docker 有個非常清楚的認識了： Docker 是一個能幫我們在各種不同 OS 上建置開發環境的工具 Docker 三元素包含 Dockerfile、Docker 映像檔（Image）以及 Docker 容器（Container） Docker 利用 Dockerfile 預先建置好一個 Docker 映像檔。在使用者想要使用容器的時候，以該映像檔為基礎，運行一個對應的 Docker 容器 Docker Hub 上有各式各樣可以直接供使用的映像檔你只需要 docker pull 及 docker run 就能開始一個分析專案給自己鼓鼓掌！現在這張 Docker 的示意圖在你眼裡應該變得平易近人許多正因為我們是資料科學家，利用 Docker能幫我們抽象化很多不必要的環境建置工作，加速我們的開發效率。在本系列文章的下篇出爐之前，我鼓勵你先下載 Docker，並開始在 Docker Hub 或者 Google 搜尋一些你感興趣的映像檔，甚至自己寫一個 Dockerfile 將你目前的專案打包起來跟別人分享。雖然我們這篇因為篇幅關係沒有細講，但只要有一個 Dockerfile，你就能使用 docker build 來輕鬆建立一個自給自足的 Docker 映像檔。一個 Dockerfile 也不難寫，像是上面貓咪的 App 的 Dockerfile 也不過就如此幾行：在本篇裡頭我們都是在自己的機器上使用 Docker。在下篇，我們將利用本篇學到的 Docker 知識，將其運用在浩瀚無垠的雲端平台之上，去最大化我們的影響力。在那之前你可以先熟悉熟悉 Docker，下次遇到你的 DS 同事時，可以問問他/她：嘿！你的 Docker Image 呢？ Originally published at leemengtaiwan.github.io.

給資料科學家的 Docker 指南

65

給資料科學家的-docker-指南-122140cd6a68

2018-09-10

2018-09-10 11:49:14

https://medium.com/s/story/給資料科學家的-docker-指南-122140cd6a68

false

885

null

Docker

docker

Docker

13,343

李孟(Lee Meng)

資料科學家，現居東京。軟體工程出身，最近努力學習資料科學並加以應用。假日在東京市區騎著腳踏車冒險、尋找美食。懶得動的時候就待在咖啡廳寫寫程式

b7c3862d5333

leemeng

36

13

20,181,104

null

0

null

0

null

2018-01-01

2018-01-01 13:35:31

2018-01-01

2018-01-01 14:08:44

1

false

th

2018-01-01

2018-01-01 14:11:22

3

12214ce838e4

0.935849

12

1

0

หมายเหตุ: บทความนี้เน้นที่ความง่าย ในการทำ Text sentimental Analysis แบบสำเร็จรูปเป็นหลัก คนอ่านจะไม่ได้รับความรู้เชิง Algorithmเลย…

4

สอนให้เครื่องจักรเข้าใจภาษามนุษย์ภายใน code 3 บรรทัด (Python — Novice Level) หมายเหตุ: บทความนี้เน้นที่ความง่าย ในการทำ Text sentimental Analysis แบบสำเร็จรูปเป็นหลัก คนอ่านจะไม่ได้รับความรู้เชิง Algorithmเลย แม้แต่น้อย…ใครต้องการอ่านในเชิงลึก รอบทความหน้านะครับ Natural Language Processing (NLP) หรือคือ การสอนให้ Machine เข้าใจภาษามนุษย์นั้นเอง ซึ่งนับว่าเป็นหนึ่งเรื่องที่ค่อนข้างท้าทายในการทำอย่างมาก เนื่องจาก ภาษามนุษย์เนี่ย มันมีความซับซ้อนค่อนข้างมาก จึงเป็นหัวข้อที่ได้รับความสนใจจากหลายคน โดยจุดประสงค์ในการทำ NLP ก็จะมีหลากหลายกันไป ตัวอย่างเช่น ให้ Machine เขียน พาดหัวข่าวให้ ให้ Machine ทำสรุปใจความสำคัญของบทความให้ ให้ Machine ดูว่า บทความนี้เกี่ยวกับหัวข้อใดบ้าง ให้ Machine ตอบ e-mail ให้อัตโนมัติ ให้ Machine detect spam mail และอื่นๆ อีกมากมายๆ แต่ที่ลืมไม่ได้เลย คือ การทำ Text Sentimental Analysis หรือการวิเคราะห์อารมณ์ ความรู้สึกจากภาษานั้นเอง โดยในการเอาไปใช้งานนั้นสามารถใช้งานได้หลากหลายไม่ว่าจะเป็น การใช้ดูแลส่วนของ Customer Service ว่าพึ่งพอใจ หรือ ไม่พึ่งพอใจ หรือ ใช้ช่วยดูแลข่าวตามพวก Social media ว่าเค้าพูดเกี่ยวกับเรา เป็น บวกหรือเป็นลบ การทำ Text Sentimental Analysis จาก 0 นั้นเป็นเรื่องที่ค่อนข้างท้าทายเป็นอย่างมากโดยเฉพาะอย่างยิ่งคนที่พึ่งหัดทำ หัดเรียนรู้ แต่หนทางก็ไม่ได้มืดมนซะทีเดียว เพราะในยุคปัจจุบันนี้ Open Souce Languange อย่าง Python เป็นที่แพร่หลาย แถมยังมีคนหลายๆคนเขียน Library มากมายในการทำงานอีกด้วย ดังนั้น การเขียน Text Sentiment Analysis แบบง่ายๆ ของเราจึงใช้ code เพียง 3 บรรทัดเท่านั้นเอง!! มาเริ่มกันเลย ก่อนที่เราจะลงไปที่เขียน code เราต้องทำการลง library สำหรับทำ NLP ภาษาไทยก่อนด้วยการใช้คำสั่งนี้ ที่หน้า terminal/cmd pip install pythainlp “pythainlpl” เป็น library ของpython ในการจัดการ NLP ภาษาไทย ใครอยากรู้อะไรเพิ่มเติมสามารถไปอ่านได้ที่นี้ pythainlp ส่วนใครขี้เกียจอ่านอยากเขียน code เลย…ตามมาทางนี้ เลย(บทความนี้เขียน code ผ่าน jupyter notebook) from pythainlp.sentiment import sentiment #เรียก library message = “สวัสดีปีใหม่2018” #สร้างข้อความ sentiment(message) #ทำการวิเคราะห์ เพียงเท่านี้โปรแกรมทำ text sentimental analysis แบบง่ายๆก็เสร็จแล้ว!! โดยผลลัพธ์ที่ออกมาจะเป็น “pos” ย่อมากจาก positive หรือข้อความเป็นบวก ในกลับกันหากผลลัพธ์ออกมาเป็น “neg” ย่อมากจาก negative หรือข้อความเป็นลบนั้นเอง ตัวอย่างโปรแกรมที่รันแล้ว ปล. Library ของ pythainlp จะคืนค่าเป็นแค่ “pos” กับ “neg” เท่านั้น หากใครต้องการทำมากกว่านี้เช่น คืนออกมากเป็น “happy” “sad” “boring” หรืออื่นๆต้องมีความเข้าใจในการทำ text sentiment เชิง algorithm ซึ่งจะยกตัวอย่างในบทความถัดไปครับ วิธีเขียนโค้ดบน Jupyter Notebook วิธีลง Python (Anaconda)

สอนให้เครื่องจักรเข้าใจภาษามนุษย์ภายใน code 3 บรรทัด (Python — Novice Level)

15

สอนให้-เครื่องจักรเข้าใจภาษามนุษย์-code-python-3-บรรทัด-novice-level-12214ce838e4

2018-06-14

2018-06-14 06:30:32

https://medium.com/s/story/สอนให้-เครื่องจักรเข้าใจภาษามนุษย์-code-python-3-บรรทัด-novice-level-12214ce838e4

false

195

null

Data Science

data-science

Data Science

33,617

DumpDataSci

A noobie data sci who need to share an experience

1ae7f932d103

dumpdatasci.th

341

2

20,181,104

null

0

defaults write org.R-project.R force.LANG en_US.UTF-8 #the world is crazy and I do not care: P var1 = c (1,2,3) class(var1) #Logical var1 = c(TRUE,FALSE) #Numeric var3 = c(5,6.31,8.2312) #Integer var4 = c(5L,1L,8L) #Character var5 = c('A','Harry',"Nathan") # you could use both single and double quotes for a character #Complex var6 = c(2+3i,4+7i) matrix(data,nrow,ncol,byrow,dimnames) #Command matrix(c(1)) #output [,1] [1,] 1 #Command matrix(c(1:5)) #output [,1] [1,] 1 [2,] 2 [3,] 3 [4,] 4 [5,] 5 #Command matrix(c(1:25),5,5) #output [,1] [,2] [,3] [,4] [,5] [1,] 1 6 11 16 21 [2,] 2 7 12 17 22 [3,] 3 8 13 18 23 [4,] 4 9 14 19 24 [5,] 5 10 15 20 25 #Command matrix(c(1:20),5,5) #output - The value repeats until all rows and columns are filled in the matrix [,1] [,2] [,3] [,4] [,5] [1,] 1 6 11 16 1 [2,] 2 7 12 17 2 [3,] 3 8 13 18 3 [4,] 4 9 14 19 4 [5,] 5 10 15 20 5 #Command - This will build a array with values from 1 to 25 in a matrix of 5 rows and 5 columns. There will be totally 4 (2*2 = 4) matrixes created. array(c(1:25),dim=c(5,5,2,2)) #Result , , 1, 1 [,1] [,2] [,3] [,4] [,5] [1,] 1 6 11 16 21 [2,] 2 7 12 17 22 [3,] 3 8 13 18 23 [4,] 4 9 14 19 24 [5,] 5 10 15 20 25 , , 2, 1 [,1] [,2] [,3] [,4] [,5] [1,] 1 6 11 16 21 [2,] 2 7 12 17 22 [3,] 3 8 13 18 23 [4,] 4 9 14 19 24 [5,] 5 10 15 20 25 , , 1, 2 [,1] [,2] [,3] [,4] [,5] [1,] 1 6 11 16 21 [2,] 2 7 12 17 22 [3,] 3 8 13 18 23 [4,] 4 9 14 19 24 [5,] 5 10 15 20 25 , , 2, 2 [,1] [,2] [,3] [,4] [,5] [1,] 1 6 11 16 21 [2,] 2 7 12 17 22 [3,] 3 8 13 18 23 [4,] 4 9 14 19 24 [5,] 5 10 15 20 25 , , 1, 3 [,1] [,2] [,3] [,4] [,5] [1,] 1 6 11 16 21 [2,] 2 7 12 17 22 [3,] 3 8 13 18 23 [4,] 4 9 14 19 24 [5,] 5 10 15 20 25 , , 2, 3 [,1] [,2] [,3] [,4] [,5] [1,] 1 6 11 16 21 [2,] 2 7 12 17 22 [3,] 3 8 13 18 23 [4,] 4 9 14 19 24 [5,] 5 10 15 20 25 #Command array(c(1:15),dim=c(5,5,2,2)) , , 1, 1 [,1] [,2] [,3] [,4] [,5] [1,] 1 6 11 1 6 [2,] 2 7 12 2 7 [3,] 3 8 13 3 8 [4,] 4 9 14 4 9 [5,] 5 10 15 5 10 , , 2, 1 [,1] [,2] [,3] [,4] [,5] [1,] 11 1 6 11 1 [2,] 12 2 7 12 2 [3,] 13 3 8 13 3 [4,] 14 4 9 14 4 [5,] 15 5 10 15 5 , , 1, 2 [,1] [,2] [,3] [,4] [,5] [1,] 6 11 1 6 11 [2,] 7 12 2 7 12 [3,] 8 13 3 8 13 [4,] 9 14 4 9 14 [5,] 10 15 5 10 15 , , 2, 2 [,1] [,2] [,3] [,4] [,5] [1,] 1 6 11 1 6 [2,] 2 7 12 2 7 [3,] 3 8 13 3 8 [4,] 4 9 14 4 9 [5,] 5 10 15 5 10 list(data) #command val1 = c("harry","nathan") val2 = c(1,4,5) val3 = c(TRUE,FALSE,TRUE) list(val1,val2,val3) #result of the list command [[1]] [1] "harry" "nathan" [[2]] [1] 1 4 5 [[3]] [1] TRUE FALSE TRUE data.frame(data) #command var1 = c(1:4) var2 = c("Harry","Nathan","John","Peter") data.frame(var1,var2) #Result var1 var2 1 1 Harry 2 2 Nathan 3 3 John 4 4 Peter #command data.frame(airquality) #Result Ozone Solar.R Wind Temp Month Day 1 41 190 7.4 67 5 1 2 36 118 8.0 72 5 2 3 12 149 12.6 74 5 3 4 18 313 11.5 62 5 4 5 NA NA 14.3 56 5 5 6 28 NA 14.9 66 5 6 7 23 299 8.6 65 5 7 8 19 99 13.8 59 5 8 9 8 19 20.1 61 5 9 10 NA 194 8.6 69 5 10 .... #command val1= c(TRUE, FALSE, TRUE, FALSE) val2= c(FALSE, TRUE, TRUE, FALSE) print(val1 & val2) print(val1 | val2) print(val1 && val2) print(val1 || val2) print(!val1) #Results > print(val1 & val2) [1] FALSE FALSE TRUE FALSE > print(val1 | val2) [1] TRUE TRUE TRUE FALSE > print(val1 && val2) [1] FALSE > print(val1 || val2) [1] TRUE > print(!val1) [1] FALSE TRUE FALSE TRUE if(expression 1) { } else if(expression 2) { } else { } val1 = 10 if(val1 > 0){ print("the value is greater than 0") } else if(val1 < 0){ print("the value is less than 0") } else print("Not resolved") switch(Expression case 1 = #Statement 1 case 2 = #Statement 2 case 3 = #Statement 3 default Statement ) #Example1 #command var1 = "10" switch (var1, '5' = print("value is 5"), '10' = print("value is 10"), '15' = print("value is 15"), '20' = print("value is 20"), print("invalid input") ) #Result [1] "value is 10" #Example2 #command switch (5, '1' = print("Monday"), '2' = print("Tuesday"), '3' = print("Wednesday"), '4' = print("Thursday"), '5' = print("Friday"), print("Invalid input") ) #Result [1] "Friday" #Example3 #command switch ("@", '1' = print("Monday"), '2' = print("Tuesday"), '3' = print("Wednesday"), '4' = print("Thursday"), '5' = print("Friday"), print("Invalid input") ) #Result [1] "Invalid input" repeat { commands if(condition) { break } } #Command var1 = 10 repeat{ print(var1) var1 = var1 + 2 if(var1 > 20){ break } } #Result [1] 10 [1] 12 [1] 14 [1] 16 [1] 18 [1] 20 while (condition){ //Statements } #command var2 = 20 while(var2 < 50) { print(var2) var2 = var2 + 10 } #Result [1] 20 [1] 30 [1] 40 for(value in vector){ //statements } #command for(x in 1:5){ print(x) } #result [1] 1 [1] 2 [1] 3 [1] 4 [1] 5 #commands and their results > var1 = "Harry" > var2 = "Nathan" > var3 = paste(var1, var2) #Concatenate two strings > print (var3) [1] "Harry Nathan" > print(nchar(var3)) #Length of a character [1] 12 > print(toupper(var3)) #To Upper cases [1] "HARRY NATHAN" > print(tolower(var3)) #To Lower cases [1] "harry nathan" > print(substring(var3,2,5)) #Substring [1] "arry" #Defining a function funcName <- function(a){ print (a) } #Calling a function funcName (5) # 5 is the parameter passed to the function #command funcA = function (a) { print (a + 1) } funcA (5) #Result [1] 6 #Command val1 = c (1,2,5,2,3,6,1) fact1 = factor (val1) print (fact1) #Result [1] 1 2 5 2 3 6 1 Levels: 1 2 3 5 6 There are two occurrences of 1, 2. The levels in the result print only unique values in ascending order

77

48e0d489b7a9

2018-09-17

2018-09-17 17:13:18

2018-09-17

2018-09-17 17:14:05

9

false

en

2018-09-17

2018-09-17 17:17:37

2

122181f87c46

8.716981

0

Hey Guys! 😃 Many of you have been mad on learning R / Python for Big data / Data science related projects. Congratulations for those who…

4

Is R really that simple? An Intro Hey Guys! 😃 Many of you have been mad on learning R / Python for Big data / Data science related projects. Congratulations for those who have already landed in their dream jobs as a Data Scientist Few students who are aspiring to be a Data scientist may have already begun to make your hands dirty by practicing R exercises. Good luck on that! R is a simple programming that was discovered in 1976 mainly used for statistical computing. What? Then why are we just discovering now to use R? I got your question on why R is prevalent now but not before? Businesses before did not have large data sets (I mean the billions of data) in their analyzes and therefore there was no need to process them. As everyone knows, we lived in the world of Excel and thought that it was almost impossible to process large records in seconds. R is the IRON MAN for Data science 💪 💪 💪 Instructions to Download and Install the Softwares STEP 1 👉 Download ‘R’ Please click on the link https://cran.r-project.org/ to download ‘R’. In case you directly install ‘R Studio’, you would get the below error. STEP 2 👉 Download the RStudio Software You go to this link to download the Open Source edition — https://www.rstudio.com/products/rstudio/download/#download If you’re Super Rich, you could also get the other versions STEP 3 (ONLY FOR MAC USERS) 👉Adjust system settings After you start ‘R’ and ‘R Studio’, if you get errors like the above, please execute the following command in mac terminal and then restart R STEP 4 👉 Sorry, there is not any! 😝 Good Job! you have installed ‘R’ Is r simple Programming languages / frameworks such as PHP, Javascript, PL / SQL are easy to learn. Similar to that, R is also a very simple programming language that you can learn quickly. The depth of knowledge lies in the amount of exercises that you work. The MORE the BETTER! ✗ Comment out a line - Each programming language has its own way of commenting out the code, In ‘R’, the special character ‘#’ is used as a single line comment. In the below articles, I would be using ‘#’ to denote the commented part if the code and comments are mentioned on the same line. Below is a simple commented line, Declaring a variable - <variable_name> = c (<value>) Examples - Identify the datatype of a particular variable Although you find can find it in “Environment” tab of R Studio, you could also see the datatype using the below command, Data types Below are the various types of variables that are used in R Programming Vector, Matrix, Array, List, Data Frame VECTORS Types : Logical, Numeric, Integer, Character and Complex. Below are the examples of each of the different types of vectors. MATRIX Matrix is an two dimensional rectangular layout Parameters data — Input vector nrow — No.of rows ncol — No.of Columns byrow — If TRUE, then elements are arranged by Row dimname — names assigned to Rows and Columns Note: All the parameters The Now lets look at an example — ARRAYS Arrays is used to store data in more than 2 dimensions Parameters data — the vector data array dimensions — first two values to describe size of each array. The multiplied value of 3rd and 4th values describes the occurrence of each array If Number of values (vector) is lesser than the Matrix size, the values will restart in the matrix which is unfilled and will start to resume from the number where it is left off in the previous matrix. Below is an example - LISTS Lists can hold different datatypes Syntax example Data Frame Two dimensional table or array like structure. Its mostly used when working with data imported from excel or any data source syntax Example — Other common pre loaded data sets can be accessed by the same command, example — Logical Operators AND, OR, NOT Using the AND and OR operator two times will only compare the first element Examples - Conditional Statements IF Statement Syntax Example Switch Case Statement Syntax Example Repeat Statement Syntax Example While Loop Syntax Example For Loop Syntax Example SEQUENCE Operators Since the syntax is fairly simple, I’ll directly explain you with the examples Functions Similar to other programming languages, a function in R is a simple reusable code that can be called any number of times with the required parameter Syntax Example FACTORS Factors print the unique occurrences of each value present in the vector. This is very important as it will be used in determining probability in statistical problems Now you have to learn the basics of R and how to use simple commands and their purpose. I’ve posted various topics / issues related to R, Data Science and Data Warehousing. Please read them as well if you are interested. Feel free to contact me in case of any doubts / issues that you have face during R coding and I would be glad to help! Thanks for reading.

Is R really that simple? An Intro

0

is-r-really-that-simple-an-intro-122181f87c46

2018-09-17

2018-09-17 17:17:37

https://medium.com/s/story/is-r-really-that-simple-an-intro-122181f87c46

false

1,992

https://github.com/harinathselvaraj

null

harinathselvaraj

null

coding&stuff

harinath.selvaraj@outlook.com

harinathselvaraj

DATA SCIENCE,DATA WAREHOUSING,BIG DATA

null

Data Science

data-science

Data Science

33,617

Harinath(Harry)

https://github.com/harinathselvaraj

af214877da6c

harithebosss

1

43

20,181,104

null

0

null

0

null

2017-12-28

2017-12-28 22:23:19

2018-01-03

2018-01-03 07:34:55

8

false

en

2018-01-03

2018-01-03 07:44:28

6

12222e9a886d

12.069182

2

1

0

The text that follows is an abridged and simplified version of the documentation vignette for my oaxaca R package. Part 1 introduced the…

5

Package “oaxaca” for Blinder-Oaxaca Decomposition in R — Part 2: Features and Example The text that follows is an abridged and simplified version of the documentation vignette for my oaxaca R package. Part 1 introduced the package and explained some econometric theory behind the Blinder-Oaxaca decomposition. Part 2, below, presents specific features of the oaxaca package, and illustrates their use with a real-world example. For a more complete treatment with full academic citations and references, please refer to the full article. If you use the oaxaca package in your research, please do not forget to cite it: Hlavac, Marek. (2018). oaxaca: Blinder-Oaxaca Decomposition in R. R package version 0.1.4. https://CRAN.R-project.org/package=oaxaca The oaxaca package consists of the main function oaxaca(), which performs the Blinder-Oaxaca decompositions, as well as of a related plot() method that produces a bar graph visualization of the decomposition results. In this section, I offer a brief overview of these functions’ capabilities. A more detailed description of the arguments and output of both functions can be obtained by typing ?oaxaca or ?plot.oaxaca into the R console. Decomposition Estimation: Main Function oaxaca() The main function oaxaca() performs both the threefold and the twofold variants of the Blinder-Oaxaca decomposition using observations from the data frame provided in the data argument. The linear regression model for the Blinder-Oaxaca decomposition is specified through the formula argument. Users can pass on a multiple-part formula that specifies the dependent variable (y), the explanatory variables (x1, x2, x3, etc.), as well as an indicator variable (z) that indicates whether an observation belongs to Group A (when z equals FALSE or 0) or Group B (when it equals TRUE or 1). These variables, along with the functional form of the model, are passed on to the formula argument in an object of class “Formula” from the Formula package by Achim Zeileis and Yves Croissant. Typically, the model formula takes the following form: y ~ x1 + x2 + x3 + … | z If the regression model contains dummies that represent a categorical variable (d1, d2, d3, etc.), these can be specified by adding another part to the formula: y ~ x1 + x2 + x3 + … | z | d1 + d2 + d3 + … When categorical variable dummies are specified, the oaxaca() function will automatically adjust estimates to be invariant with respect to the user’s choice of the omitted baseline category. If the user does not include any other arguments, oaxaca() will estimate the Blinder-Oaxaca decompositions — both threefold and twofold — based on Ordinary Least Squares regressions (estimated via the standard lm() function), and will calculate standard errors based on 100 bootstrapping replicates. By default, oaxaca() estimates the twofold decomposition with Group A coefficients, Group B coefficients, their equally weighted average, a weighted average that reflects the number of observations in Groups A and B, as well with pooled coefficients — both including and excluding the group indicator variable — as the set of reference coefficients. These defaults can, however, easily be changed. Users can use the argument group.weights to specify additional relative weights of Group A and Group B coefficients in the estimation of the twofold decomposition. They can also choose, via the R argument, how many bootstrapping resamples should be drawn to calculate the standard errors. Last but not least, users can use a different regression function (argument reg.fun) to estimate the regression coefficients used in the decompositions. Note that, if a non-linear function such as glm() is chosen, the decomposition will be based on the linear systematic component — usually associated with the estimation of the corresponding latent variable — of the regression method. The function oaxaca() returns an object of class “oaxaca”, which can then be passed on to the plot() method to obtain a bar graph visualization of the Blinder-Oaxaca decomposition results. The object contains lists named threefold and twofold which contain the results of the threefold and twofold decompositions, respectively. In addition, the object stores the regression coefficients used in the decomposition (component beta), the number of observations in each group that were used in the analysis (n), the number of bootstrapping replicates (R), the regression objects generated during the analysis (reg), as well as the mean values of both the dependent variable (y) and the explanatory variables (x). Visualization: Method plot() The oaxaca package can produce easily customizable bar charts that visually summarize the results of its Blinder-Oaxaca decompositions. All bar charts are generated using the ggplot2 package by Hadley Wickham. To visualize the decomposition results, the user simply passes an “oaxaca”-class object created by the main function oaxaca() to the plot() method. Users can choose which of the estimated decompositions to visualize. The decomposition argument determines whether a threefold or a twofold Blinder-Oaxaca decomposition will be shown, while the type argument specifies whether the bar graph will contain an overall decomposition or a detailed, variable-by-variable one. If the detailed decomposition type is selected, component.left determines whether decomposition components or variable names will be aligned along the left side of the graph. The argument weight allows the user to select which of the twofold decompositions should be shown, and the unexplained.split argument determines whether the unexplained components ought to be split into the two discrimination subcomponents (“unexplained A” and “unexplained B”). Users can, furthermore, choose which of the variables and decomposition components will be shown (arguments variables and components), as well as their labels (variable.labels and component.labels). Standard error bars that indicate confidence intervals can be toggled by the ci argument, and the confidence level adjusted by ci.level. Several formatting options are available. The bar graph’s title can be set using the title argument, and axes can be labeled by xlab and ylab. Finally, users can change the colors of the bars by specifying the bar.color argument. Example: Wages of Native and Foreign-Born Workers In this section, I use an empirical example to demonstrate the capabilities of the oaxaca package. In particular, I use the Blinder-Oaxaca decomposition to explain the wage gap between native and foreign-born Hispanic workers in metropolitan Chicago. I analyze data from the chicago data frame, included in the oaxaca package: R> data(“chicago”) The chicago data frame contains information about the demographic characteristics and labor market outcomes of 712 employed Hispanic workers in the Chicago metropolitan area. It is a subset of the 2013 Current Population Survey (CPS) Outgoing Rotation Groups (ORG) data set. I am interested in decomposing the wage gap between native and foreign-born workers. The wage gap could be due to group differences in the level of wage determinants such as age, gender or education. Alternatively, the gap could arise from a differential effect of these determinants on native and immigrant workers’ wages. I call the oaxaca() function to estimate the relative magnitudes of these channels’ influence: R> results <- oaxaca(formula = real.wage ~ age + female + LTHS + some.college + college + advanced.degree | foreign.born | LTHS + some.college + college + advanced.degree, data = chicago, R = 1000) As the formula argument indicates, the outcome variable in this decomposition is real.wage, the worker’s real wage denominated in 2013 U.S. dollars. The values of the dependent variable had been obtained by exponentiating the natural logarithm of the workers’ real wages (contained in the provided ln.real.wage variable): R> chicago$real.wage <- exp(chicago$ln.real.wage) The linear regression model includes covariates that account for the workers’ age, gender and education. LTHS (“less than high school”), some.college, college and advanced.degree are indicator variables that denote the highest level of education an individual has achieved. A high school education is the omitted baseline category. The variable foreign.born indicates whether a worker was born outside of the United States. Group A consists of native workers, and Group B of foreign-born ones. To make sure that the choice of the omitted baseline does not affect the decomposition estimates, the formula argument also specifies that the categorical variables denoting the education level ought to be adjusted. Bootstrapped standard errors are calculated based on 1,000 replicates. R> results$n $n.A [1] 287 $n.B [1] 379 $n.pooled [1] 666 The n component of the resulting “oaxaca”-class object indicates that there are 287 native and 379 foreign-born workers in the analyzed sample. The pooled analysis contains 666 observations. R> results$y $y.A [1] 17.58282 $y.B [1] 14.56725 $y.diff [1] 3.015574 The y component of the resulting “oaxaca”-class object indicates that the mean real wage is $17.58 for the natives (Group A) and $14.57 for foreign-born workers, leaving the difference of approximately $3.02 to be explained by the Blinder-Oaxaca decomposition. Threefold Decomposition First, I look at the results of the threefold Blinder-Oaxaca decomposition: R> results$threefold$overall The results of the threefold decomposition suggest that, of the $3.02 difference, approximately $1.62 can be attributed to group differences in endowments (i.e., age, gender, education), $2.83 to differences in coefficients, and the remaining -$1.43 is accounted for by the interaction of the two. Next, I examine the endowments and coefficients components of the threefold decomposition variable by variable. This is most easily done by using the plot() method: R> plot(results, components = c(“endowments”,”coefficients”)) Figure 1 shows the estimation results for each variable, along with error bars that indicate 95% confidence intervals. In the endowments component, most variables appear to have a statistically insignificant (or only marginally significant) influence, with the sole exception of LTHS. It seems that a significant portion of the native-immigrant wage gap is driven by group differences in the proportion of individuals with less than a high school education. Figure 1: The endowments and coefficients components of a threefold Blinder-Oaxaca decomposition of the native vs. immigrant wage gap. R> summary(results$reg$reg.pooled.2)$coefficients[“LTHS”,] R> results$x$x.mean.diff[“LTHS”] LTHS -0.2693959 Individuals with less human capital tend to earn less, as can be seen from the pooled regression coefficient on LTHS reported above. Furthermore, the value of x.mean.diff shows that a greater proportion of foreign-born Hispanic workers have not attained a high school education. The difference in the educational composition of native and immigrant worker groups thus accounts for some portion of the natives’ higher wages. Similarly, most variables are either insignificant or exhibit only marginal statistical significance in the coefficients component. The only variable which achieves clear statistical significance is age. R> results$beta$beta.diff[“age”] age 0.1860063 As the difference in the age coefficients between natives and immigrants shows, the wage payoff of an additional year of age is greater for U.S.-born Hispanic workers by almost 19 cents. As Figure 1 makes clear, differences in the regression coefficients on age account for the decisive portion of the wage gap. Twofold Decomposition Next, I look at the results of the twofold Blinder-Oaxaca decomposition. In the output below, the group.weight column indicates the relative weights of coefficients from a regression on observations from Groups A and B, respectively, in the reference coefficient vector. The two negative weights indicate that the reference coefficients come from pooled regressions either without (-1) or with (-2) the group indicator variable included as a covariate. R> results$twofold$overall For presentational ease, I focus my discussion on the decomposition that uses pooled regression coefficients (from a regression that does not include the group indicator variable foreign.born) as the reference coefficient set. This decomposition is denoted by -1 in the weights column. The results of the overall twofold decomposition indicate that the $3.02 wage gap between native and foreign-born Hispanic workers can be decomposed into $1.36 that can be explained by group differences in the explanatory variables and $1.66 that is unexplained. Let us assume that the unexplained component of the wage gap occurs due to labor market discrimination, and that the pooled regression coefficients are non-discriminatory. The Blinder-Oaxaca decomposition would then also indicate that $0.94 of the unexplained part originates from discrimination in favor of native Hispanic workers (component “unexplained A”), while $0.72 comes from discrimination against those who are born outside of the United States (component “unexplained B”). The standard errors provide a sense of the uncertainty that accompanies all of the point estimates. R> plot(results, decomposition = “twofold”, group.weight = -1) Figure 2 provides a variable-by-variable twofold decomposition. The results are consistent with the threefold composition. It appears that the wage gap is driven largely by the lower proportion of workers with less than a high school education among the natives (in the explained component) and by the native workers’ greater returns to age. Figure 2: The explained and unexplained components of a twofold Blinder-Oaxaca decomposition of the native vs. immigrant wage gap. I can explore the unexplained component even further. In Figure 3, I examine three variables from the decomposition — age, female and college — and visualize how much of the unexplained portion of the wage gap can be attributed to discrimination in favor of the natives, and how much is due to discrimination against the immigrants. R> plot(results, decomposition = “twofold”, weight = -1, unexplained.split = TRUE, components = c(“unexplained A”, “unexplained B”), component.labels = c(“unexplained A” = “In Favor of Natives”, “unexplained B” = “Against the Foreign-Born”), variables = c(“age”, “female”, “college”), variable.labels = c(“age” = “Years of Age”, “female” = “Female”, “college” = “College Education”)) I use a variety of plot() method arguments to customize the formatting of the resulting bar graph. Through the components and component.labels arguments, I choose to display only the two subparts — “unexplained A” (i.e., discrimination in favor of Group A) and “unexplained B” (discrimination against Group B) — of the unexplained decomposition component, and attach appropriate labels to them. Similarly, I use the variables and variable.labels arguments to select and label the variables I examine. Here’s the result: Figure 3: The unexplained portion’s discrimination sub-components in a twofold Blinder-Oaxaca decomposition of the native vs. immigrant wage gap. It appears that only the discrimination components for the age variable (labeled “Years of Age” in the bar graph) achieve non-marginal statistical significance. The relative size of the bars suggests that — if we assume that the pooled regression coefficients reflect a state of non-discrimination — almost twice as much of the wage gap is explained by discrimination against foreign-born workers as it is by discrimination in favor of native ones. The comparison would be a little easier to make if the discrimination components bar charts were presented side-by-side for each variable separately. This can be achieved by switching on the component.left argument in the plot() method. The resulting bar graph is presented in Figure 4. R> plot(results, decomposition = “twofold”, weight = -1, unexplained.split = TRUE, components = c(“unexplained A”, “unexplained B”), component.labels = c(“unexplained A” = “In Favor of Natives”, “unexplained B” = “Against the Foreign-Born”), component.left = TRUE, variables = c(“age”,”female”,”college”), variable.labels = c(“age” = “Years of Age”, “female” = “Female”, “college” = “College Education”)) Figure 4: The unexplained portion’s discrimination sub-components in a twofold Blinder-Oaxaca decomposition of the native vs. immigrant wage gap. An alternative presentation. Specific numerical values of the point estimates of the unexplained discrimination components can, of course, be obtained directly from the “oaxaca”-class object: R> variables <- c(“age”, “female”, “college”) R> columns <- c(“group.weight”, “coef(unexplained A)”, “coef(unexplained B)”) R> results$twofold$variables[[5]][variables, columns] To summarize, I have used the Blinder-Oaxaca decomposition to examine the wage gap between native and foreign-born Hispanic workers in the Chicago metropolitan area. The results of my analysis suggest that much of the gap can be explained by two facts: There are more workers with less than a high school education in the foreign-born group. Workers with a lower stock of human capital tend to command lower wages in the labor market. As a result, the relatively less-educated group of foreign-born Hispanic workers will, on average, earn lower wages than their native counterparts. The returns to age are greater for native workers than for the immigrants. In other words, even if the foreign-born workers had the same average age as the natives, the native group would, on average, earn higher wages than immigrants. This result makes some intuitive sense if we interpret age as potentially picking up the effect of labor market experience. The higher returns to age among the natives may, for instance, reflect the differential availability of more lucrative jobs with greater opportunities for career growth. In this two-part article, I have introduced the oaxaca package for the R statistical programming language. It is the first R package that allows researchers to estimate Blinder-Oaxaca decompositions, a statistical method that decomposes differences in mean outcomes across two groups into a part that is due to group differences in the levels of explanatory variables and a part that is due to differential magnitudes of regression coefficients. oaxaca estimates threefold and twofold Blinder-Oaxaca decompositions for linear models, and also provides estimates for a detailed, variable-by-variable decomposition. Each point estimate is presented with a bootstrapped standard error that measures the corresponding estimation uncertainty. Part 1 introduced some of the theory behind these decompostion. Here, in Part 2, I have demonstrated the package’s capabilities through an empirical example that examines the wage gap between native and foreign-born Hispanic workers in the Chicago metropolitan area. In doing so, I have also showcased the oaxaca package’s unique visualization features that allow users to graphically summarize the results of their decompositions. About the author: Marek Hlavac teaches economics at UWC Adriatic, an international secondary school in Duino (Trieste), Italy. In addition, he is a research fellow at the Central European Labour Studies Institute (CELSI) in Bratislava, Slovakia. He holds a Bachelor’s degree in Economics from Princeton University, a Master’s degree in Public Policy from Georgetown University, and a Ph.D. in Political Economy and Government from Harvard University. You can find his personal website here.

Package “oaxaca” for Blinder-Oaxaca Decomposition in R — Part 2: Features and Example

2

package-oaxaca-for-blinder-oaxaca-decomposition-in-r-part-2-features-and-example-12222e9a886d

2018-04-11

2018-04-11 10:06:57

https://medium.com/s/story/package-oaxaca-for-blinder-oaxaca-decomposition-in-r-part-2-features-and-example-12222e9a886d

false

2,898

null

Data Science

data-science

Data Science

33,617

Marek Hlavac

Economics Teacher at UWC Adriatic in Duino, Italy; Research Fellow at the Central European Labour Studies Institute (CELSI) in Bratislava, Slovakia

bcf2a64a4cae

MarekHlavac

110

266

20,181,104

null

0

null

0

aa381d373b60

2017-11-16

2017-11-16 12:53:31

2017-11-16

2017-11-16 12:54:09

0

false

en

2017-11-16

2017-11-16 12:54:09

2

122284539212

1.735849

0

Every day we read reports pointing out how crucial time wastage and poor resources allocation, along with operational and procedural…

2

Clinical Artificial Intelligence: a New Approach to Healthcare Every day we read reports pointing out how crucial time wastage and poor resources allocation, along with operational and procedural factors, lead to inefficient healthcare systems. Health professionals and patients recurrently encounter inefficiency in terms of the extra burden of needless additional diagnostic tests, false positives leading to misdiagnoses, and mistreatment, which all culminate into a number of avoidable situations: delayed treatment, poor survival rate, and soaring remission rate. Do we have a way to optimize healthcare? Well, the answer to the question — clinical artificial intelligence — seems to be kind of drawn from a science fiction; however, it certainly offers a way capable to bring a paradigm shift in the healthcare industry. Modern AI-based healthcare solutions have capability to identify disease, or at least aid diagnosis, track medicine response, help chart treatment plan, and automate a barrage of mundane tasks, which come with health care. Evidently, the computer-aided self-learning technology can reap benefits the industry as well as patients. According to McKinsey, healthcare sector, along with pharmaceutical industry, can annually save up to $100 B with smarter decisions. Gauging the opportunity, a number of blue chip companies and a barrage of starts-ups has started to capitalize the capability of the AI, machine learning, and healthcare data analytics. We are at a stage where analytics and machine learning are poised to bring paradigm shift in the healthcare industry: Disease Diagnosis The most of research in clinical artificial intelligence and machine learning is focused on ways to change diseases are diagnosed. We are seeing ways where integrated systems, with ability to read and understand medical data, assisting health professionals to economically diagnose diseases, such as cancer, heart diseases, and other life threatening health problems, and monitor conditions like jaundice, diabetes, hyperthyroidism, lung functions, and hemoglobin. Evaluation of Disease at DNA level Machine learning algorithms have given us ability to evaluate and analyze copious amount of data and draw insights, which were one almost impossible to draw. By evaluating the genome of individuals, with the help of predictive analytics, personalized healthcare can be given. With the application of machine learning, healthcare would reach a level conceptualized in science fictions. Better Treatment Options It is noticed that most care-giving problems stem from delay in relaying information and miscommunication, and AI can eliminate all such factors. AI based healthcare solutions are also affecting treatment options and tracking of the efficiency of a medicine, as well as assisting patients to adhere to the treatment plan better. With the help of AI, health professionals now can find the latest treatment option by crowd sourcing and can learn about the patients better and effectiveness of the medications and treatment in the real time.

Clinical Artificial Intelligence: a New Approach to Healthcare

0

clinical-artificial-intelligence-a-new-approach-to-healthcare-122284539212

2018-05-14

2018-05-14 12:13:37

https://medium.com/s/story/clinical-artificial-intelligence-a-new-approach-to-healthcare-122284539212

false

460

Jvion’s RevEgis solution is an award winning predictive solution to predict the risk of an illness before symptoms occur.

null

Jvion

null

jvion

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

jvionhealth

null

9a3e55f8076d

jvionhealth

2

1

20,181,104

null

0

null

0

null

2018-09-01

2018-09-01 23:33:23

2018-08-30

2018-08-30 12:50:06

1

true

en

2018-09-01

2018-09-01 23:33:48

3

12251cfdf0ae

1.633962

0

When first researching the topic of Augmented vs. Virtual Reality for SIX Marketing, 2018 Keith couldn’t help but flashback to 1998 Keith…

4

Augmented Vs. Virtual Reality: Why You Should Give a S*@t! — Single-life, Sales, Sports, Sobriety, Suicide, & Screenwriting When first researching the topic of Augmented vs. Virtual Reality for SIX Marketing, 2018 Keith couldn’t help but flashback to 1998 Keith thinking Augmented Reality was like doing a hit of clean LSD and Virtual Reality was comparable to eating an 1/8th of potent psilocybin mushrooms. (May I say I nailed the spelling of psilocybin on the first attempt? Which is as awesome as it is pathetic.) Virtual Reality is total immersion. Submerged in an alternate reality. A world which you have little control of. You’re at the behest of a designer. A demigod. You’re a marionette doll, and the pleeb with a degree from MIT and Redcloak level in Dungeons and Dragons is your puppet master. (Take note, to be clever, I had to Google the highest level of D&D, so kiss my ass!) Augmented Reality is enhanced reality. It’s subtle, gradual, like looking through a pair of dirty sunglasses. Instead of the lenses being covered in nasal mucus and snot as if a friend just sneezed on them, they are covered with Pikachus and Instagram face filters. (Yes, for those of you who have no goddamn clue what Augmented reality, it’s the filter you use on match.com and Snapchat. Which by the way, makes you all look sexy as hell. I never knew I wanted to have sex with a bunny.) The primary and most important difference between the two is this: Accessibility. Instead of having to go out and excitedly buy a pair of glasses which will hang off your unused Bowflex in a month (VR). All you have to do is use the phone you can’t stop looking at every other fucking second. (AR) Oh, and the fact the same dipshits who named these VR glasses are the same assholes who named dating sites! (Oculus, Zoosk, Christian Mingle). So, before you go out and spend your pizza delivery tips on a pair of VR glasses, read this blog post I did for SIX Marketing. It will be quite revealing. And if your a business owner too, read it, it may very well be the future of marketing. Enjoy, -k Originally published at www.shortbitteritalian.com on August 30, 2018.

Augmented Vs.

0

augmented-vs-12251cfdf0ae

2018-09-04

2018-09-04 16:11:18

https://medium.com/s/story/augmented-vs-12251cfdf0ae

false

380

null

Virtual Reality

virtual-reality

Virtual Reality

30,193

Keith Hannigan

null

f0de00419287

keithhannigan

1

11

20,181,104

null

0

null

0

null

2018-03-30

2018-03-30 07:36:22

2018-03-30

2018-03-30 07:49:33

1

false

en

2018-03-30

2018-03-30 07:49:33

8

1225c28c87a3

3.286792

1

0

A news item regarding missing children caught my attention recently.

3

Face recognition software for tracking missing children. A news item regarding missing children caught my attention recently. https://timesofindia.indiatimes.com/city/delhi/missing-piece-of-puzzle-face-matching-software-can-aid-parents-search-for-kids/articleshow/63358470.cms Kailash Satyarthi’s NGO Bachban Bacho Andolan (BBA) has been pleading that all Child Care Institutions in India should be integrated with the Governments, Track The Missing Child portal and face recognition software should be used to locate missing children quickly. This was an excellent use case in the AI4Good space and motivated me to explore this topic. This article is an attempt to provide a birds eye view of where we are and how we got here. The intention is to encourage students to explore this use case using open source tools. Apple iPhone X Face ID technology is the most visible example of the current state of the art in this field. This white paper provides some details. https://images.apple.com/business/docs/FaceID_Security_Guide.pdf Following points are noteworthy. - The technology is centered around neural networks. - The neural engine is run on Apple’s A11 Bionic Processor. - The neural network was trained using over a billion images. - iPhone X uses TrueDepth camera system comprising of IR emitter and IR camera to create an image and depth map of the face. - This data is sent to the neural network to create a mathematical representation of the face which is then compared to the mathematical representation of the face originally enrolled by the user. - If there is a match, a positive ID is made. - Apple claims that this technology has a 1 in a million chance of making a wrong identification compared to 1 in 50,000 for touch ID. Very impressive indeed. This technology is proprietary to Apple, but surely others like Google, Facebook, Baidu and many others will soon catch up. Let us now try to see, how this field has evolved over the years and what is the situation on the open source front. Work on face recognition software has been going on since the 1960’s. However use of deep learning for facial recognition is comparatively recent and can be traced back to the pioneering work done by Yann LeCun in the area of Convolution Neural Networks (CNN). Yann LeCun was a student of Geoffrey Hinton who is considered the father of neural networks. In 1998 Yann LeCun published his famous paper on LeNet-5, a 7 layer CNN that was used to classify digits. Since then CNNs have been widely used in computer vision and image processing. Yann LeCun later became the director of Facebook AI Research Lab (FAIR) and it is only appropriate that a breakthrough paper on use of CNNs for face recognition was published by a research team at FAIR in 2014. This was called DeepFace and you can access the paper here. https://research.fb.com/wp-content/uploads/2016/11/deepface-closing-the-gap-to-human-level-performance-in-face-verification.pdf DeepFace was trained on a dataset of 4.4 million images and reported an accuracy of 97% compared to 85% for the FBI Next Generation Identification System at that time. Further improvements were then suggested by a research team at Google who proposed a CNN based approach called FaceNet in 2015. FaceNet paper can be accessed here. https://arxiv.org/pdf/1503.03832.pdf FaceNet was trained on a dataset of 200 million images and reported an accuracy of 99.63 %. By now it should be obvious that the internet giants with access to very large training datasets had a lead over others and their models were proprietary. Fortunately research teams at several universities like Oxford and CMU have helped to publish models in public domain which are not far behind. For example CMU lab has published its work under the name of OpenFace and can be accessed here. http://cmusatyalab.github.io/openface/ Similarly the Visual Geometry Group (VGG) at Oxford publishes it’s work here. http://www.robots.ox.ac.uk/~vgg/ FaceNet implementation using Tensorflow is available here. https://github.com/davidsandberg/facenet OpenFace uses Torch but for those who prefer Keras there is also a Keras version of OpenFace available here. https://github.com/iwantooxxoox/Keras-OpenFace Since I started by saying that I would like to encourage students to explore face recognition using open source tools, here is my suggestion. Learn Python, Tensorflow and Keras. All of this is in public domain and there are a lot of online learning tools. The Keras OpenFace face recognition model referenced above provides an accuracy of 93.8% and can give good results. Here is how the solution will work for locating missing children. - All children in child care institutions will be photographed and pictures will be uploaded into the track the missing child portal. - Pictures of all missing children are uploaded to the track the missing child portal. - Track the missing child portal runs all pictures through the face recognition CNN and generates the feature vectors of each picture. - Feature vectors of missing children pics are matched with feature vectors of pics of children in child care institutions and an alert is generated for each match. Happy Learning !!.

Face recognition software for tracking missing children.

1

face-recognition-software-for-tracking-missing-children-1225c28c87a3

2018-03-30

2018-03-30 17:59:57

https://medium.com/s/story/face-recognition-software-for-tracking-missing-children-1225c28c87a3

false

818

null

Machine Learning

machine-learning

Machine Learning

51,320

Sudhir Gupta

null

db575b06f919

sudhir.g

1

20,181,104

null

0

null

0

null

2018-05-18

2018-05-18 07:39:46

2018-05-18

2018-05-18 07:50:13

1

false

en

2018-05-18

2018-05-18 07:50:13

3

1225e7639fa7

2.14717

0

When 60 researchers — some of the most intelligent people on the planet — band together for a single cause it’s time to listen. That’s what…

4

4 Modern Life Lessons Learnt from OpenAI’s Elon Musk When 60 researchers — some of the most intelligent people on the planet — band together for a single cause it’s time to listen. That’s what happened with the formation of OpenAI. What is OpenAI? Spearheaded by the well known inventor and entrepreneur Elon Musk this body’s goal is to ensure the development of safe AGI. Yes, apparently it’s not absurd thinking robots could take over the world. Even Stephen Hawking stated if AI is given the chance to reinvent itself it could backfire on humanity. Luckily plans are in place — such as what’s happening at OpenAI — to ensure AGI (Artificial General Intelligence) always has a positive effect on humanity. This is done by researching possibilities and being proactive about creating conditions in which AGI can be safely developed. This forward thinking approach is only one thing we should all learn from Mr. Musk. His success can be attributed to his intelligence but also his wisdom about life in general. Here are a few examples. Technology Should Always Benefit Humanity This relates directly to the motivation behind OpenAI. In a world driven by profit and power one of the most influential people on earth still thinks of the impact tech has on humanity. With this approach driving manufacturing, business and tech development there will be less abuse of technology to benefit a single person, bank account or country. It’s Okay to Have Fun But Elon Musk is definitely not serious all the time. You must have a sense of humor if you decide to strap your car to a rocket and then play rock music during its launch. So apart from involving himself in new inventions Elon also found out how to combine different aspects of his life. Perhaps this ability to balance them — instead of compartmentalizing them — is part of his keys to success. Social Media Can Be Used for Good Here’s one more fact that proves his balanced outlook on life. Instead of only sitting in his office focusing on making more money this business mogul actually cares about politics, economy and more. Elon’s Twitter messages often mention relevant topics that are of value to the average person. His informative statements and opinions can give you a unique perspective to help gain insight on important matters. And then there’s of course the lighter side of life. My favorite Elon Musk tweets are his thoughts on inventions society needs to simplify life. Don’t Ignore Criticism But here’s the most impressive thing about this guy. You’ll find it on a TED talk and it made me think about how I value people in my life. When asked about the reason for his extreme success Elon attributed some of it to his decision to listen to people around him. When people advise him, criticize him or share their opinions, this guy listens. You would think a business leader of his calibre would only trust himself. If what he says is true I had better start listening more to those around me. What about you?

4 Modern Life Lessons Learnt from OpenAI’s Elon Musk

0

4-modern-life-lessons-learnt-from-openais-elon-musk-1225e7639fa7

2018-05-18

2018-05-18 07:50:14

https://medium.com/s/story/4-modern-life-lessons-learnt-from-openais-elon-musk-1225e7639fa7

false

516

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

C Sutton

C Sutton is a passionate freelance copywriter based in Johannesburg, South Africa. His skills involve SEO copywriting & user-friendly blogging.

15ca650d2f60

ablogondating

0

1

20,181,104

null

0

null

0

null

2018-02-09

2018-02-09 19:38:15

2018-02-04

2018-02-04 14:47:20

2

false

en

2018-02-09

2018-02-09 19:41:17

7

12266aef0b14

2.866352

4

0

This post is part of x.ai’s ‘Future of Jobs’ interview series. We’re talking to leaders and mavericks to find out how emerging technologies…

5

Pondering AI’s future with Azeem Azhar This post is part of x.ai’s ‘Future of Jobs’ interview series. We’re talking to leaders and mavericks to find out how emerging technologies like AI are changing how they do their jobs. You can find all the posts in the series here. Named a LinkedIn “Top Voice” for the past two years, Azeem Azhar has established himself as a vital voice at the intersection of technology and society. Founder of the venture-backed startup PeerIndex, the entrepreneur, investor, and former reporter currently publishes the essential weekly newsletter Exponential View. We had a delightful exchange with Azhar about the current divide between human and AI problem solving, and his hopes for automation in the future. Let’s talk about you first, Azeem, what aspects of your job would you like to see automated? What aspects of your job are already being automated or will be, in your view, over the next few years? Improvements in prioritisation would be helpful, as there are too many opportunities and too much information. I’d love AI to be able to triage more efficiently to reduce what comes across my limited attention window. One part of my job, which is about exploring the space of the possible, has already benefited from automation. Web search engines and filtering tools have replaced much flicking through index cards at a library. What parts of your job are you uniquely qualified to do (vs. AI, machine learning, etc.)? I enjoy listening to people, asking them questions and learning from them, which will be hard for AI to learn to do genuinely. I also enjoy thinking about thinking and other aspects of metacognition. So I’m deliberate about trying other approaches to thinking through a problem, like analogic thinking, five WHYs, or applying a lens from different disciplines. I’m not sure AI will do that yet. If artificially intelligent assistants could take over some of the tasks you’re used to doing, what would your day look like? What would you spend that extra time doing? My problem solving falls into a number of categories: go deeper (to the most granular I can manage), go broader (to emphasis analogical problem solving), collaborate on a tricky issue, talk to someone (to socialise an idea or drive action), talk to an expert (to learn from them) or do something different (to let an issue foment in the back of my mind). If I’m not problem solving, then I need time to build the tree of actions and follow-up with whoever is doing It. In other words, any time bonus will be put to good use! What’s the main thing you wish you had more time for? It doesn’t have to be related to work. I would want to travel with my family and explore more of the world with them, especially while my kids are young. If AI gave you time back, would you spend it pursuing recreational interests or would you spend it doing more work? It is hard to know whether that would be the effect. It might be that I’d just to do more work. I’d like to think that I’d become like the fisherman in this Paulo Coelho parable but I’m not sure if many of us have the discipline to do that. If you accept the premise that AI will fundamentally change your job in some way over the next decade, what will your job look like then, and what qualities will drive your success? I’ve never done a job that existed the decade before I did it. My first job — as tech correspondent of the Guardian in 1994 — involved [coding] a Web publishing system and covering the Netscape IPO. I don’t look at my work over a ten-year time horizon. Too much changes too quickly. Want to hire Amy + Andrew? Start your free trial HERE Lead image via widewalls Originally published at x.ai on February 4, 2018.

Pondering AI’s future with Azeem Azhar

26

pondering-ais-future-with-azeem-azhar-12266aef0b14

2018-06-10

2018-06-10 18:00:59

https://medium.com/s/story/pondering-ais-future-with-azeem-azhar-12266aef0b14

false

658

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

x.ai

Magically schedule meetings

2795062a44df

xdotai

7,963

1,327

20,181,104

null

0

null

0

379cf85b825e

2018-04-29

2018-04-29 21:00:56

2018-05-07

2018-05-07 18:01:01

2

false

en

2018-05-07

2018-05-07 19:54:12

18

1226db7739e0

6.168239

6

0

Behold the AI overlords of your fitness future

5

BotBoxer I Attempted to Fight A Robot Faster Than Floyd Mayweather Behold the AI overlords of your fitness future When I first heard about a punching bag that could dodge punches — any punches, no matter how fast — I laughed. First because of how exciting it was to hear that such technology could exist, but then because I seriously doubted whether it could be any good. It’s not that I consider myself the second coming of Muhammad Ali, but it’s fair to say that I’ve hit a fair amount of punching bags in my life, including shifty ones designed to confuse your aim and get you off your rhythm. Plus, over the last century, little has changed in the equipment people use to train for striking sports like boxing or Muay Thai. The tools of the trade have always been brutally elegant in their simplicity — gloves and a bag, primarily — and I wondered if BotBoxer, which claims to sense your movement and dodge blows faster than any human can, could live up to its hype. I certainly wasn’t alone in my skepticism. Tony Jeffries, the pro and Olympic boxer who won bronze for England at the 2008 Summer Games, first spotted the contraption at a conference for the International Health, Racquetball and Sportsclub Association held in San Diego in March. He was there to rep his gym, Box N Burn, and noticed chatter around him about a robotic punching bag capable of eluding haymakers of every kind and speed. Jeffries, himself endowed with a sneaky-fast jab, couldn’t resist trying to prove otherwise. “If there’s an exciting new development in the boxing industry, I’m the first one there, and I’m also the biggest critic,” he explains. “There are so many gimmicks, so much BS out there. I’ve got a very nice car, you know, and I would’ve bet the thing that I could hit this machine.” So Jeffries sauntered up to the BotBoxer booth and discovered an unusual-looking contraption. Designed by the company SkyTechSport, BotBoxer sports a punching “bag” that sits on top of an arm that’s connected to a heavy motorized base, making it look like an oversized joystick. (SkyTechSport’s first simulator was an ingenious ski/snowboard contraption that motors you across a horizontal rail in front of a screen to mimic the effects of charging down a snowy path.) Metal tubing forms a half-circle cage around the sides and rear of the BotBoxer, offering about a 180-degree area that you can attack. To the right is a touchscreen monitor with myriad available settings, including reaction time, how far the bag can dodge and even the probability of successful dodges. Jeffries squared up, stared the machine down and snapped a jab out. Whiff. He squared his shoulders again and shot another jab out. Whiff again. “I couldn’t hit the thing,” Jeffries says, grinning. “This machine is faster than Floyd Mayweather, I really believe that. When I finally hit it after lowering the reaction time, I felt a little better about it.” He’s telling me all of this from his Santa Monica gym, which hosted the machine last month and where I’ve been invited to try it. Milidzhan Gevorkyan, who oversees product marketing for SkyTechSport, warms me up by asking me to shadowbox in front of the machine. Next, he has me inch closer and closer to it, the point being to calibrate my perception of how sensitive it is to movement. The subtlety of its movement catches me off guard: When I wiggle my lead shoulder, feinting a jab with my right hand, the bag wiggles to my left. If I square my torso to it with a step forward, it jerks back and settles into a lean. I start throwing some easy punches, and quickly discover that BotBoxer isn’t using jerky, unnatural mechanical movements to dodge my shots — it’s slipping them by a centimeter or two, fluidly snapping back to center after I whiff. Having warmed up, Gevorkyan suggests I battle the Botboxer on its “Fight” setting, which attempts to replicate some of the human qualities you might find in a live sparring session. “If you keep it moving, it’ll start to tire. And if it tires, you can try landing some bigger combos,” he explains. “If you can get it from 100 percent down to zero, it’ll be knocked down.” For the first minute, I’m confident I’m going to murder this thing. I’m moving light on my feet, attacking the red bag with quick jabs and left-hand crosses that seem to land consistently, though I’m not getting much power into them. Still, the BotBoxer is already down to 65 percent strength. Feeling emboldened, I start throwing larger, looping shots with my left hand. The thrill of catching the bot hard on consecutive hits gives me a surge of giddiness — at least until I realize how exhausted I’m becoming. Boxing wisdom observes that a whiffed punch is more tiring than one that lands, and I’m whiffing a lot. Frustrated, I begin to throw more long shots, but they’re accompanied by sloppy footwork that makes me jerk sideways. The final 30 seconds of the three-minute round leaves me second-guessing every shot in a bid to save energy. (“Go! Faster! It’s regaining energy!” Gevorkyan urges). I’ve done a decent job, but fallen short of my goal of a first-round knockdown. Also: I’m officially dripping in sweat. Overall, among the most challenging aspects of striking training is understanding how to manipulate distance between you and an opponent. Without actually sparring a real human or working hand pads with a coach, it’s extremely difficult to replicate the small movements that determine whether a left cross lands or misses. There are plenty of examples of how calculated dodging can ruin an average person’s ability to hit anything at all, and getting knocked out in the ring is often a byproduct of leaning in too much, or losing balance on a wild, reaching punch. The BotBoxer attempts to account for such minutiae. The machine uses high-speed motion-tracking cameras to read the body in front of it, and three servomotors in the black plastic base move the punching bag swiftly and quietly on four axes. “So whenever you punch it, it can read the movements of your shoulders and see if you’re pulling your shoulder back to load up a strike or try to throw a faster one,” project leader Alex Golunov tells me. “The cool part is that allows you to train more deceptive punching, without so much telegraphing, which you want to avoid in a real boxing situation.” Currently, SkyTechSport is trying to get BotBoxers into boxing gyms with the hope of courting more feedback from fighters, but the ultimate goal is to sell the machines to mainstream gyms. The $20,000 price tag may look daunting, but considering top gyms can spend upwards of $10,000 on modern treadmills, Golunov sees major market potential, noting that a lot of people get bored in gym routines and could benefit from the rig’s competitive, game-like drills and fight settings. Market studies would seem to back him up as the mixed martial arts training industry is primed to grow rapidly in the next four years, with high rates of return expected from punching bags in particular. There are some unintentionally hilarious examples of other robotic boxing prototypes, but to Golunov, it’s more a matter of safety than technology and basic mechanics. For instance: While BotBoxer took more than four years to develop, the vast majority of that time was spent engineering it to be consistently safe. “We could injure people if it tracked a body in the background and moved the wrong way,” Golunov explains. So for now at least, this little machine stands as a realistic glimpse into the future of working out. Jeffries, for one, has become a convert. “I believe anything is possible now,” he says. “What if this became a sparring partner where you can pick height, punch power and length? If someone makes a machine that could hit you back, that would be amazing for the industry. Because then you really can box Floyd Mayweather every day.” Eddie Kim is a staff writer at MEL. He last compiled a gentleman’s guide to cooking for a date. Most popular stories on MEL: When You Can and Cannot Say ‘No’ When Someone Asks to Switch Seats With You on an Airplane For seven years, Tim Roscoe worked as a fund-raiser for the University of Virginia, taking upwards of 20 flights per…melmagazine.com The Three Things Women Want You to Say in Bed… Daddy At least, according to possibly questionable survey results from an Australian diamond companymelmagazine.com A Guy Walks Into a Bar — and Is Never Seen Again On April Fools’ Day, a med student went for a drink. A security camera recorded him. Then, he vanished.melmagazine.com Death by Muscle Rich Piana is the latest in a long list of untimely deaths. What’s sending bodybuilders into early graves?melmagazine.com The Gentleman’s Guide to Paying a Bar Tab If you’ve never tended bar before, the process of ordering, consuming and paying for drinks warrants little thought…melmagazine.com

I Attempted to Fight A Robot Faster Than Floyd Mayweather

16

i-attempted-to-fight-a-robot-faster-than-floyd-mayweather-1226db7739e0

2018-05-19

2018-05-19 23:49:44

https://medium.com/s/story/i-attempted-to-fight-a-robot-faster-than-floyd-mayweather-1226db7739e0

false

1,533

There's no playbook for how to be a guy

null

melmagazine

null

MEL Magazine

editor@melindustries.com

mel-magazine

MEN,LONGFORM,MAGAZINE,JOURNALISM,LIFESTYLE

wearemel

Boxing

boxing

Boxing

4,161

Eddie Kim

LA-based reporter penning features for Mel Magazine.

16edac58890d

eddiekimx

668

219

20,181,104

null

0

null

0

null

2017-12-05

2017-12-05 13:29:31

2017-12-05

2017-12-05 14:54:18

1

false

fr

2017-12-08

2017-12-08 15:06:17

0

12281a4962a8

2.283019

1

0

(et du monde)

5

Google : Le Futur Boss de l’Intelligence Artificielle (et du monde) Cette illustration n’a aucun rapport avec l’article, j’aime juste beaucoup Tim & Eric ❤ Pour faire une bonne IA, on a besoin de deux choses : d’un bon logiciel & de beaucoup de données. Beaucoup de données pour pouvoir entrainer et alimenter le logiciel et un bon logiciel pour pouvoir les interpréter et prendre des décisions. Google a les deux et d’une façon qui dépasse de loin toute concurrence. (Pour ce que l’on connaît du moins…). Des logiciels d’Intelligence Artificielle parmi les meilleurs du monde Deux exemples qui sortent des labos de Google : AlphaGo qui a battu les meilleurs joueurs humains à un jeu réputé inaccessible à l’IA et AlphaGo Zero qui a battu quelques temps après AlphaGo avec une nouvelle méthode d’apprentissage et sans le recours à l’historique des parties humaines. Waymo qui développe un logiciel permettant à une voiture de rouler en toute autonomie. Selon les dernières données disponibles, l’IA de Waymo est 10 fois plus performantes que son plus proche concurrent. Google le roi indétrônable de la donnée C’est peut-être là que Google a l’avance la plus nette : Google a dès le départ eu comme ambition d’organiser toute la donnée du monde. Voici quelques exemples de données dont dispose le géant de Mountain View : Gmail : première plateforme d’email pour 1 Milliards d’utilisateurs à travers le monde (et par extension la suite Google Drive et les services associés) Google : le moteur de recherche indexe (et dans bien des cas mêmes héberge) la quasi totalité des contenus présents sur Internet : textes, images, vidéos… A une part de marché de 88% de la recherche à l’échelle mondiale et accumule des quantités massives de données sur les recherches et les comportements des internautes. Google Chrome : 64% de parts de marché dans le monde (c’est titanesque), le navigateur permet là encore à Google de connaître l’ensemble du comportements des internautes. Androïd : 82% de part de marché et première OS pour téléphone mobile, permet de récupérer des données de géolocalisation et comportementale. Google Maps / Earth : cartographient l’ensemble de la terre en 3 dimensions en mêlant prises de vue terrestres et par satellite et peuvent récolter les données de géolocalisation des personnes utilisant ses services. Google book : a pour objectif de numériser l’ensemble des livres publiés sur terre et s’attelle à cette tâche depuis 2004. Google Analytics : La quasi totalité des sites ont Google Analytics en place, ce logiciel récolte les données de visite et de comportement de l’ensemble des internautes sur Internet. (Temps de visite, dépenses etc…). Google AdWords & Adsense : Première régie publicitaire au monde et en ligne, récolte elle aussi de nombreuses données sur les comportements et les performances publicitaires. Mais aussi des activités dans la biotechnologie, la domotique et bien d’autres encore sur lesquelles la maison mère ne communique pas et qui sont regroupées dans la société Google X. A côté Facebook et Amazon avec leurs énormes stocks de données qualifiées font figures de lilliputiens de la data. On ne parle même pas des acteurs européens… S’il doit y avoir un Skynet, ce sera Google. A tout le moins ce sera l’un des acteurs majeurs de l’Intelligence Artificielle. Les gains de productivité et la valeur que l’IA créera ira droit dans la poche de Google monstre tout puissant de notre futur.

Google : Le Futur Boss de l’Intelligence Artificielle

1

google-le-futur-boss-de-lintelligence-artificielle-12281a4962a8

2018-03-14

2018-03-14 16:23:43

https://medium.com/s/story/google-le-futur-boss-de-lintelligence-artificielle-12281a4962a8

false

552

null

Google

google

Google

35,754

Cosmo

https://twitter.com/abc_0_xyz

9d9741bf48b5

abc_0_xyz

91

367

20,181,104

null

0

null

0

null

2017-09-21

2017-09-21 13:23:40

2017-09-21

2017-09-21 15:00:03

3

false

en

2017-09-21

2017-09-21 15:00:03

1

1228bdaeb98

2.414151

0

Yet another one story about analytics. But this one will be different. This story will not teach super advanced way how to calculate CLTV…

3

Google Analytics “session” explained for dummies. Yet another one story about analytics. But this one will be different. This story will not teach super advanced way how to calculate CLTV or ROMI. This read will try to explain what is one simple metric — sessions. And me, author, will try to make sure you will have the easy time reading it. Varom. Things you must know before reading it: What is a browser What is a webpage Who is the author of this golden post If you are reading this (if at least someone is reading this) you should know what is a pageview in Analytics. For those who don’t know — it’s one-page load by a user. So if you grandma goes to www.recipes.com/kugel Google Analytics will count it as one pageview for kugel recipe page. Simple. Every page load from a user is a pageview by a user. In hard words — a page is a dimension, pageview — a metric. “Dimension??? Wth, is it the matrix or what?” Nah, it’s Google’s dimension. It’s like an X on XY chart. Dimension in Google Analytics is like x line in xy chart For those who had idea what pageview is, and now stopped believing the story I’m telling. You are right and I’m wrong. I have described a pageview from data point of view not from techinacal, where pageview is a simple pageload. If you read this you are one of 20% who started reading this or one of my relatives (usually half of those 20%). Let me tell you another metric which is few times more complex — session. In simple words, a session is all pageviews done by a user in one browsing session (that’s why it’s called sessions probably). So if you brows Page1->Page2->Page3 without going to have a coffee, you had a session of three pages. Why did I add a coffee break to my sentence? Because Google Analytics identify sessions by timeouts between pageviews. All it scumbag cares is a time between pageviews. If you opened page1 and after short time page2 it’s a single session. But if you had a longer break, most probably, GA will count it as a new session. Clarifying “most probably” it’s 30 minutes timeout, but you can adjust it to whatever you like. To make things harder I have to say that, sessions are totally another dimension than pageviews. Aww, ugly truth. Sessions is not a dimension, it’s a metric, but it can’t be combined together with pages. “Hypocrite! It’s made of pageviews, why I can’t combine it reports?”. Because, as I said, it’s different dimension, layer, level call it what ever you like. Now I will leave you here. I won’t go deeper into details because it will make your head and my heart hurt. What I have said take as facts and only true. Remember that session is made of pageviews (or events if you have any). And compute it in your head. See you next time, peace.

Google Analytics “session” explained for dummies.

0

google-analytics-session-explained-for-dummies-1228bdaeb98

2018-01-19

2018-01-19 13:44:41

https://medium.com/s/story/google-analytics-session-explained-for-dummies-1228bdaeb98

false

494

null

Google Analytics

google-analytics

Google Analytics

3,111

Mantas A.

null

de6ad525d250

Andriuska

19

18

20,181,104

null

0

null

0

e50bb4537c90

2018-06-01

2018-06-01 13:12:53

2018-06-01

2018-06-01 13:49:54

2

false

en

2018-07-10

2018-07-10 09:17:41

2

122923d0a1ce

4.315409

0

Zkylos Updates & Transforms The Pet Society

5

AI [For Animals] They Said? (Whatever Next!) Zkylos Updates & Transforms The Pet Society For those of you who don’t know, it is about time you got to grips with the 21st century AND GOT TO KNOW. Because technology is here and bigger than we could have ever imagined. Apart from the fact that we are not in a flying car quite yet . . . The dawn is upon us for great revolutionary CHANGE. Society (as we know it) has already evolved and continues to transform. Statistics reveal that humans are, in fact, quite a lonely bunch — so much so, that we like to keep ourselves amused with a trusty companion or pet. We have about 54 millions of pets globally — according to PFMA (pfma.org). The trouble is, acquiring a pet and living aside your loved one is not as easy as one might think . . . It is not as straightforward as traipsing to the pet store and carrying home a suiter safe and sound, no no no. Forget what you fathomed in your imaginary world of fluffy bunnies dancing around rainbows. RELATED: Flaws In The System Episode 3 Paperwork? These Guys Are ‘Kidding!’ Whichever family member you decided is for you — whether that be a dog, cat, horse — whatever! (Something you probably didn’t wholeheartedly vouch for) exactly just how much lifetime paperwork you are setting yourself up for. (Not to mention the copious amounts of time waiting for the damn offices to get back to you with their approval . . . To then tell you-you made a damn typing error and effectively, you must start all over again and pay the exact same fee twice over!) The Question Is: WHY?? AND . . . Are we not in the new millennium? Do we not have the means to rectify most situations with computers and simplify operations to optimize convenience? Can we not yet use technology to our advantage at this stage of evolution? WTF are we still doing filling in PAPERwork? (In an age of computers? There is a goddamn reason this is a method that is called — just that — ‘work’ !!!) Furthermore . . . What lies around the corner once this mess is cleared up? How far can we stretch the capacity of convenience? What happens when we use machines to our advantage? Imagine what can be achieved! The sky’s the limit! THANK THE HEAVENS A digital angel came along and blessed the Earth, and its hard-working beings with the grace of Zkylos. (You can laugh all you want, but the truth is A LOT of people and animals are going to benefit from this platform on a global scale.) That’s right — this is no joke. You know something? The best ideas often get shunned, to begin with: “Why do we need that?” “Are you serious?” Well actually, yes, we are. Very. You see, if Zkylos can boost this market (in the way that it so longs to be catapulted, from a prehistoric era of meddling with a variety of offices) — of which don’t even communicate — and allow for the community to freely interact and provide a way to share data properly . . . THEN there becomes this window of opportunity to escalate possibilities. So Let’s Talk About Those Possibilities! If we incorporate Artificial Intelligence into an implemented system that functions based on data input, we create amazing potential and ease-of-use. Essentially, Zkylos has invented purpose and value for the following: “Perfect Match” Indicator Mating and breeding [animals] (lol) has never been easier! It might sound like something similar to that of an ‘animal dating site,’ but in actuality, this function will liberate means to choose and set specific criteria. And why not? Animals are social creatures too, you know! Wait for it . . . !!! Yes, this means now you can decide the colors! Eyes, coat and you can even determine the body type of your pet. What is the world coming to? Seriously though — with machine assistance limitations are few. You can pretty much design the litter (if you were working with dogs for example). Genetic Deviation Predictor Now here is where we get truly technical. Scientists [and the likes] now have access to a better insight into the health history of animals with the ability to predict the future for species. A provided percentage of probability of an animal’s descendants leads the way to eradicate potential health risks. You know what this means, right? In the grand scheme of things, we have acquired tactic to take control of wellness and therefore, prolong the existence of animate beings at risk of extinction. Zoos and sanctuaries that work hard to take great care of beloved creatures that are low in numbers can now identify issues far easier and find the qualities in procreation that prevents the decrease further. Pattern Identification Once again, a reason to show off! A function that alerts breeders of rare genetics. Again, in the name of Science researchers can detect exact correlations and consistencies between specifically set occurrences related to pets. Game Changer As you can see, some next level actions take the role of social and agricultural responsibility that would take far longer for humans to manually accomplish and things like identification require a non-biased regulation. The neutral transparency provided by machine assistance maintains real authenticity that people, pet-owners, breeders, scientists and regulators (as well as the rest) can 100% rely upon. No guessing games are inflicted in this bad boi. RELATED: “The ICO market changed,” they said… To Sum Up YES! AI is necessary for animals; there is no doubt about it. A service that is as transparent as is decisive with an approach that is more exact and precise than any mortal attendee or operator is an impressive step forward for the entire pet society as a whole. If you have an opinion on the matter that you would like to share, please do not hesitate to comment in the box below. We are more than grateful to hear your thoughts, answer your questions and hear any suggestions that you may have! (It is a free world, after all.)

AI [For Animals] They Said? (Whatever Next!)

0

iai-for-animals-they-said-whatever-next-122923d0a1ce

2018-07-10

2018-07-10 09:17:41

https://medium.com/s/story/iai-for-animals-they-said-whatever-next-122923d0a1ce

false

1,042

Digital Animal Identity

null

zkylos

null

Zkylos

null

zkylos

null

ZkylosZKS

Zkylos

zkylos

Zkylos

23

Diana

null

c7b1a73d5d74

nikonovadiana

1

2

20,181,104

null

0

null

0

null

2018-06-18

2018-06-18 07:56:12

2018-06-18

2018-06-18 07:58:47

2

false

en

2018-06-18

2018-06-18 07:58:47

1

122a5b81146b

2.304088

0

Are you wondering about the hype around Data Science? Data says that it’s not a hype anymore. Glassdoor released its Report of 50 Best jobs…

1

Gearing up for a career in Data Science? Don’t forget the Intangibles Are you wondering about the hype around Data Science? Data says that it’s not a hype anymore. Glassdoor released its Report of 50 Best jobs in America in 2017. With a job score of 4.8 out of 5, a job satisfaction score of 4.4 out of 5, and a median base salary of $110,000, Data Scientist jobs came in first for the second year in a row. IBM predicts demand for Data Scientists will soar 28% by 2020. Undoubtedly, a career in Data science is a lucrative one. Data science is an interdisciplinary field which can become a career option for most aspirants with appropriate training and self-motivation. Usually, the pre-requisite is an educational degree in any of the following disciplines — mathematics, computer science, economics, statistics, engineering, operations research or business management. However, aspirants from other educational streams can also get into this field with appropriate training. What is required is a dedicated focus on learning the hard skills that enable you to first get selected in a data science job and then perform and subsequently develop your career in data science. These hard skills include, but are not limited to, programming and data mining capabilities through languages like R, Python, SAS, SQL, Hive, Pig, Julia, SPSS etc as well as deep understanding of statistical techniques for data analysis and analytical solutions like basic descriptive statistics, hypothesis testing, regression and segmentation techniques and some exposure to advanced techniques like machine learning and optimization. The web is flooded with detailed information and guidance on these. Analyttica TreasureHunt could well-become a partner in this journey of yours, watch out for this platform. In this race to get up-skilled and apply for positions that match the hard skill requirements, aspirants tend to overlook the “X-factors” that often turn out to be the deciding parameter for job selection. These factors, the Intangibles, are usually the unspoken qualities that employers look for in aspirants, when interviewing them for a data science role. The weight on these intangibles often increases with increasing years of experience. While you cannot gauge the exact questions around the intangibles during an interview, you can prepare for various situations beforehand and embed some of these into your day-to-day — this will help you become a stronger data science professional in due course. In this blog, we try to give you a snippet of five of the most important intangible skills, in an effort to kick-start your preparation for that job you have been eyeing in data science. 1. Investigative Mind Curiosity and the hunger to ask questions fearlessly mostly make a great analyst . The ability to constantly investigate existing hypothesis and create new ones, often defines the key responsibility of many data science jobs. Data scientists are expected to spot changes in trends and identify the cause behind the change. This requires a strong eye for detail along with the tendency to associate causality with changes. Read more

Gearing up for a career in Data Science? Don’t forget the Intangibles

0

gearing-up-for-a-career-in-data-science-dont-forget-the-intangibles-122a5b81146b

2018-06-18

2018-06-18 07:58:48

https://medium.com/s/story/gearing-up-for-a-career-in-data-science-dont-forget-the-intangibles-122a5b81146b

false

509

null

Data Science

data-science

Data Science

33,617

David

null

e678c334e1d2

David.435

1

20,181,104

null

0

null

0

1f35b6f451e8

2018-08-03

2018-08-03 07:52:50

2018-08-03

2018-08-03 12:19:51

0

false

en

2018-08-03

2018-08-03 12:19:51

0

122b1c27ad44

0.056604

0

null

5

Webography for 4 dummies to make it in machine learning — Chapter 23, Scene 5

0

webography-for-4-dummies-to-make-it-in-machine-learning-chapter-23-scene-5-122b1c27ad44

2018-08-03

2018-08-03 12:19:52

https://medium.com/s/story/webography-for-4-dummies-to-make-it-in-machine-learning-chapter-23-scene-5-122b1c27ad44

false

15

We offer contract management to address your aquisition needs: structuring, negotiating and executing simple agreements for future equity transactions. Because startups willing to impact the world should have access to the best ressources to handle their transactions fast & SAFE.

null

ethercourt

null

Ethercourt Machine Learning

adoucoure@dr.com

ethercourt

INNOVATION,JUSTICE,PARTNERSHIPS,BLOCKCHAIN,DEEP LEARNING

ethercourt

How To Make It

how-to-make-it

How To Make It

266

WELTARE Strategies

WELTARE Strategies is a #startup studio raising #seed $ for #sustainability | #intrapreneurship as culture, #integrity as value, @neohack22 as Managing Partner

9fad63202573

WELTAREStrategies

196

209

20,181,104

null

0

null

0

null

2018-03-31

2018-03-31 01:31:02

2018-04-05

2018-04-05 16:08:48

1

false

en

2018-04-09

2018-04-09 04:51:51

17

122cf87113a2

5.837736

0

A Debate Over the Dangers of AI

1

An Artificial Future A Debate Over the Dangers of AI With a single comment made over Facebook Live, Mark Zuckerberg thrust himself in to the center, if only for a moment, of what is becoming a very important and hotly debated topic. Right now we are racing towards a world where artificial intelligence (AI) manages all of the standard processes in our world, eliminating many human labor jobs. But AI could be taking over more than just our jobs; it could take over our society without us realizing it, if we don’t take caution when developing it. At least- that’s what Elon Musk believes. However, Zuckerberg has a less concerning image of the future, imagining one in which AI is integrated in to our society only in order to improve human lives. In a Facebook Live Q&A, Zuckerberg disregarded the calls for caution about AI as “irresponsible,” and shared his opinion with the audience that regulation might hinder the growth of the emerging technology. The two entrepreneurs are both attempting to take the lead on AI research, but are approaching the issue from very different perspectives. Considering the contrasting views these business leaders hold, it is worth considering how valuable their opinions are. Background and Experience Mark Zuckerberg has been an avid computer programmer since he was very young. One of his first major applications was a messaging system used by his family known as “Zucknet”. Later in life, Mark famously dropped out of Harvard to start Facebook in 2007 at just 20 years old. Over those last 10 years, Zuckerberg has lead the company as CEO, driving the company’s mission to “give people the power to build community and bring the world closer together.” As the leader of a multibillion dollar tech company, Zuckerberg realized early the need for constant innovation and created the Facebook Artificial Intelligence Research (FAIR) team. Since its inception, FAIR has become a world leader in AI research by utilising Facebook’s large databases of contextual information, produced by its user, to construct complex neural networks that attempt to mimic human learning patterns. Elon Musk has a very different background from Mark Zuckerberg in both subject breadth and depth. Musk was born and raised in South Africa until age 17 when he move to Canada to avoid mandatory Military Service. He pursued his education at Queen’s College, and later completed his Bachelors in Economics at the University of Pennsylvania. Always the dedicated learner, Musk decided to stay and complete a second Bachelor's degree in Physics before moving to California to get a Masters in Energy Physics at Stanford. He soon abandoned his plan of pursuing the graduate degree in favor of starting his first company, Zip2, which is best described as an early version of Yelp. After Zip2 sold for over $300 million, Musk became a serial entrepreneur starting x.com (later paypal.com) and SpaceX, as well as buying a controlling interest in an existing company, Tesla. With his accumulated wealth from these ventures, Musk has since join forces with fellow entrepreneurs Peter Thiel, Sam Altman, Reid Hoffman, Greg Brockman, and Jessica Livingston to create a non profit known at OpenAI. OpenAI is dedicated to the “discovery and enacting the path to safe artificial general intelligence (AGI)” by conducting “fundamental and long-term research.” At this point, I believe it is important that we take a moment to discuss the difference between AI and AGI. AI, in a limited sense, is already all around us. There’s the machines that conduct trades on the stock exchange in fractions of a second, better known as High Frequency Trading, which accounts for 50% of all trades. There’s also the pricing algorithms used by many online retailers to automatically update pricing based on changes in cost of inventory, demand for products, or competitor pricing. We often unknowingly interact with AI every day, but these forms of AI are limited in their abilities. They might be able to refine their own ability to perform a specific task, but its designed to be limited to those specific tasks. As for AGI, it’s essentially two steps further along. Where as a traditional AI program could learn how to perform a task better through the use of a feedback loop, an AGI would also be able to learn new tasks, and make decisions about what tasks to learn. It’s not hard to think about how this could go wrong for humans, which is a major concern of OpenAI and its founders. Given these different definitions of AI, it may be that Elon Musk and Mark Zuckerberg are talking about two different things. The Evidence Zuckerberg seems to like to use the new features on Facebook to communicate directly with his users, with the added benefit of showing them how the new feature can be used. In his Facebook Live stream, Zuckerberg sits in his San Francisco back yard with his wife. Just before taking a break from the conversation to stoke the flames of the grill, he takes a question about Elon Musk and his concerns over AI. He takes a moment to note the importance of continuing the development of AI, due to the many application is has to improve our lives. “In the next 5 to 10 years AI is going to deliver so many improvements to the quality of our lives. If you think about just safety and health, and keeping people safe, AI is already helping us diagnose diseases better, match up drugs with people so they can get treated better. So it’s going to help a whole lot of people get healthcare who wouldn’t have had access before. If you look at self-driving cars, they are going to be safer than people-driving cars, that’s only a matter of time.” — Mark Zuckerberg, Facebook Live stream July 2017 Zuckerberg is pointing to the different applications of AI, which he believes are apparent and important, in order to defend continuing research and testing it in the field. But, as we discussed above, these application are limited to the specific task that they are design to assist with. This is the AI we interact with in our everyday lives as well as every time we log in to Facebook. But it is far from the the sentient AGI that Musk is concerned about. When Zuckerberg’s charge towards Musk was pointed out to him, he was quick to dismiss it unequivocally. Musk has long urged researchers and global governments to take caution with developing AI, going as far as calling it “our greatest extinction threat” in an interview with an MIT student at the AeroAstro Centennial Symposium. After being an early investor in Deepmind, later acquired by Google, Musk partnered with Peter Thiel, Sam Altman and others to create the OpenAI project. His initial investment in Deepmind was only to allow him to keep track of AI’s growth trajectory, according to Musk himself. However, the OpenAI project was his first true effort to play a role in the development of this new technology. Right now, OpenAI is conducting research for the long term future. They’re working with what’s known as ‘Reinforced Learning AI,’ which can learn to perform different actions simply by examining the actions available to it, getting better over time. This form of AI technology is currently considered to have the greatest chances of leading to an AGI. To drive their mission “to build safe AGI, and ensure AGI’s benefits are as widely and evenly distributed as possible,” Musk partnered with Greg Brockman, a researcher from MIT, to serve and Chief Technology Officer. Other Perspectives It clear that Musk and Zuckerberg have very different perspectives on AI and its future prospects, but they both have many supporters in the research field, and across social media. Reddit has the most robust conversations about this issue. One username Hell_Libertine dismissed Musk. “Elon Musk is a generalist, not an expert by any means. He’s a public figure and as such, when combined with how AI has been portrayed in popular culture, can cause a ruckus about how AI is really dangerous, and he’s gladly done so.” Hell_Libertine also quotes Andrew Ng, Chief Scientist of Baidu and Adjunct Professor at Stanford University. Ng claims that concern about AGI and killer robots are “an unnecessary distraction.” Reddit users are known for being vocal and thorough in these debates, and the rebuttals were equally well supported. Bullockhouse pointed out that “plenty of leading experts are concerned and want research into safe control paradigms for general-purpose agents to begin now,” referencing and article that discussed Stephen Hawking and Bill Gates common concern with Musk, and about the dangers of AI in the future. While it is currently uncertain what the true dangers of AI may be, there are many people who feel that the potential is high enough that as a society, we need to consider strict standards for AI, and should continue pushing for further research in the field.

An Artificial Future

0

the-dangers-of-artificial-intelligence-122cf87113a2

2018-04-09

2018-04-09 04:51:51

https://medium.com/s/story/the-dangers-of-artificial-intelligence-122cf87113a2

false

1,494

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Drew Greene

null

188d97a19318

d.greene2701

0

1

20,181,104

null

0

null

0

null

2018-08-13

2018-08-13 12:01:20

2018-08-13

2018-08-13 12:40:16

2

false

en

2018-08-13

2018-08-13 12:40:35

6

122e1cb9c3f9

2.707862

1

0

Recap From Day 034

5

100 Days Of ML Code — Day 035 100 Days Of ML Code — Day 035 Recap From Day 034 In day 034, we looked at working with audio input: Common audio features. We learned that Spectral Centroid tells us something about the timbre of a sound. Specifically, it gives us information about how bright a sound is. Visually, we can understand Spectral Centroid by imagining we have a frequency spectrum made out of solid object. The Centroid lies just under the center of mass of this object. Today, we’ll continue from where we left off in day 034. Working With Audio Input: Common Audio Features Continued Constant Q Transform “The constant-Q transform transforms a data series to the frequency domain. It is related to the Fourier transform and very closely related to the complex Morlet wavelet transform.” “The transform can be thought of as a series of logarithmically spaced filters fk, with the k-th filter having a spectral width δfk equal to a multiple of the previous filter’s width” If you want information about pitch or timbre, but the peak frequency and Spectral Centroid don’t give you enough information, there is a nice middle ground in between those very simple features and using the full set of FFT magnitude values. Source The Constant Q transform gives us a nice, general-purpose feature vector in the middle ground between peak frequency and Spectral Centroid. Like the FFT, Constant Q gives us information about the strengths of the different frequencies present in our analysis window. However, Constant Q gives us logarithms spacing between these frequencies which sounds complicated until you realize that our human perception of frequency is basically logarithmic. We can set up Constant Q to give us one frequency bin per octave, for example. And this can give us a low-dimensional feature vector that tells us something useful about the relative lowness or highness of the sound. Or, we could use Constant Q to give us twelve bins per octave, essentially matching one bin to one musical semitone. Even if we did this for all 88 keys on the piano, this is a much more compact feature vector than, say, a 512-bin FFT. But it’s still incredibly musically meaningful. Comparison with the Fourier transform In general, the transform is well suited to musical data, and this can be seen in some of its advantages compared to the fast Fourier transform. As the output of the transform is effectively amplitude/phase against log frequency, fewer frequency bins are required to cover a given range effectively, and this proves useful where frequencies span several octaves. As the range of human hearing covers approximately ten octaves from 20 Hz to around 20 kHz, this reduction in output data is significant. The transform exhibits a reduction in frequency resolution with higher frequency bins, which is desirable for auditory applications. The transform mirrors the human auditory system, whereby at lower-frequencies spectral resolution is better, whereas temporal resolution improves at higher frequencies. At the bottom of the piano scale (about 30 Hz), a difference of 1 semitone is a difference of approximately 1.5 Hz, whereas at the top of the musical scale (about 5 kHz), a difference of 1 semitone is a difference of approximately 200 Hz. So for musical data the exponential frequency resolution of constant-Q transform is ideal… It’s good to know that you’re still here. We’ve come to the end of day 035. I hope you found this informative. Thank you for taking time out of your schedule and allowing me to be your guide on this journey. And until next time, remain legendary. Reference https://www.kadenze.com/courses/machine-learning-for-musicians-and-artists-v/sessions/sensors-and-features-generating-useful-inputs-for-machine-learning https://en.wikipedia.org/wiki/Constant-Q_transform

100 Days Of ML Code — Day 035

4

100-days-of-ml-code-day-035-122e1cb9c3f9

2018-08-13

2018-08-13 12:40:35

https://medium.com/s/story/100-days-of-ml-code-day-035-122e1cb9c3f9

false

616

null

Music

music

Music

174,961

Jehoshaphat Abu

A polymath, an advocate of STEAM education. I write about Music | Computing | Design and maybe life and the world in general

62d9f8742a1e

jehoshaphatia

189

319

20,181,104

null

0

null

0

null

2018-07-26

2018-07-26 04:27:29

2018-07-26

2018-07-26 04:41:59

0

false

en

2018-07-26

2018-07-26 04:41:59

1

122e1ff01cda

2.022642

0

The technological developments are aimed at simplifying the business operations and overall life of the common people living in the world…

1

Speech Recognition Program For Virtual Assistance In Typing, Browsing & More The technological developments are aimed at simplifying the business operations and overall life of the common people living in the world. That can be the symbol of success of the companies and individuals performing continual research to develop a solution that creates a lasting impact on the society. Artificial intelligence for simplifying life: There has been a lot of research in the area of Artificial Intelligence and most of the solutions derived through it have become the cause of revolutionary changes in the diversified industries that are using them. Artificial Intelligence Program for simple task of typing as per the dictation is now available online at the exclusive online service that brings this revolutionary solution at affordable price. How it actually works? The aspirant professionals, entrepreneurs and any other users that don’t wish to rely on a person for typing the day to day business communications and reports; can download speech recognition software for pc and get it installed to begin with. This revolutionary program would recognize the voice of the user. Thus this program can be utilized as an aid to type in as per the dictation. The aspirants would not need a physically working assistant to carry out these routine tasks as the virtual assistant with great level of efficiency would be available at the service. Diversified applications and benefits: The speech recognition program can be used for different routine tasks performed through the personal computers. Once this program is installed; the users can easily carry out the following tasks: • They can go on speaking out the matter to be included in any report. They can get it typed through the program without any need to even touch the keyboard. This can be a boon to the users that might be slow in typing speed sue to lack of habit and any other reason. The program can cope up with normal to faster speed of speech very effectively. Thus even the users may not need to take pauses while they dictate the lengthy write-ups and letters. • The speech recognition program can also be used to open the personal e-mail account that is assigned initially. Thus the users can save their time in accessing the mails and starting a business conversation through e-mail with a prospect, existing client, business ally or a supplier as well. • The users can make effective utilization of speech recognition software for windows 10 to open the desired website. Thus they can easily navigate the browser to open the sites for digging out more information for their business purposes; communication with the other key people working on the projects and for many other reasons for which the internet can be used. Availability in different languages: The speech recognition program is fortunately not restricted to English. Though the world speaks English and it has emerged as the global languages; there are many regions across the globe where some other languages are commonly spoken and understood. The users from different geographic locations and backgrounds would be able to use this program as it is offered in 13 different languages spoken in various parts of the world.

Speech Recognition Program For Virtual Assistance In Typing, Browsing & More

0

speech-recognition-program-for-virtual-assistance-in-typing-browsing-more-122e1ff01cda

2018-07-26

2018-07-26 04:42:00

https://medium.com/s/story/speech-recognition-program-for-virtual-assistance-in-typing-browsing-more-122e1ff01cda

false

536

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Wade Program

null

f1bf6f688132

wadeprogram1

0

1

20,181,104

null

0

null

0

null

2018-07-19

2018-07-19 11:54:16

2018-07-19

2018-07-19 11:55:13

1

false

en

2018-07-19

2018-07-19 11:55:38

20

122f82a1a11f

0.871698

1

0

Artificial intelligence (AI), sometimes called machine intelligence, is intelligence demonstrated by machines, in contrast to the natural…

2

AI: from idea to action Artificial intelligence (AI), sometimes called machine intelligence, is intelligence demonstrated by machines, in contrast to the natural intelligencedisplayed by humans and other animals. In computer science AI research is defined as the study of “intelligent agents”: any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals.[1] Colloquially, the term “artificial intelligence” is applied when a machine mimics “cognitive” functions that humans associate with other human minds, such as “learning” and “problem solving”.[2] The scope of AI is disputed: as machines become increasingly capable, tasks considered as requiring “intelligence” are often removed from the definition, a phenomenon known as the AI effect, leading to the quip, “AI is whatever hasn’t been done yet.”[3] For instance, optical character recognition is frequently excluded from “artificial intelligence”, having become a routine technology.[4] Capabilities generally classified as AI as of 2017include successfully understanding human speech,[5] competing at the highest level in strategic game systems (such as chess and Go[6]), autonomous cars, intelligent routing in content delivery network and military simulations.

AI: from idea to action

1

ai-from-idea-to-action-122f82a1a11f

2018-08-01

2018-08-01 11:34:20

https://medium.com/s/story/ai-from-idea-to-action-122f82a1a11f

false

178

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Innovation Services

Management Consulting Firm Based in Athens

3159544348ae

inn.servic

0

1

20,181,104

null

0

null

0

null

2018-02-28

2018-02-28 06:33:00

2018-03-09

2018-03-09 06:42:43

1

false

en

2018-04-02

2018-04-02 14:44:40

3

1232da18b07c

3.241509

2

0

2015 was the year that I started to hear the buzz around the “Sexiest Job Alive, Data Scientist”. At the time I was working as a Data…

5

Photo by Chris Ried on Unsplash Data Analyst to Data Scientist 2015 was the year that I started to hear the buzz around the “Sexiest Job Alive, Data Scientist”. At the time I was working as a Data Analyst on a big data migration project. My client was migrating some of their data processes from a legacy platform to a cloud environment, whilst at the same time creating a different stream of processes for a new product line they were launching imminently. A bit of background about myself. I studied Financial Mathematics with dreams of making millions in the investment banking world. I had a warped perspective of the world in terms of career, education, and ambition. As a graduate fresh out of university with a Mathematics degree (first class), my peers and I thought the world was our oyster. We were so wrong, I shall elaborate more on a different blog post. Fast forward to 2015, after a stint in Finance as a quantitative analyst for a trading finance which opened my eyes to the world of finance. A few unfortunate circumstances with the firm lead me to pursue a career in consulting as a Data Analyst/Consultant. Personally, having a job title as a scientist was the coolest thing ever. After discovering it involved a hybrid discipline of Computer Science, Mathematics & Statistics, I was even more sold that I wanted to become a Data Scientist. I really enjoyed working with data as an analyst and also enjoyed the coding and engineering aspects I was exposed to infrequently. THE COVERT OPERATION Personally, having a job title as a scientist was the coolest thing ever. After discovering it involved a hybrid discipline of Computer Science, Mathematics & Statistics and Business Knowledge, I was even more sold that I wanted to become a Data Scientist. I really enjoyed working with data as an analyst and also enjoyed the coding and engineering aspects I was exposed to infrequently. All I had to do now was to come up with a covert operation of enhancing my skill set from an analyst to a scientist. I had done a little bit of programming as an Analyst (SQL, VBA, C#) but my research showed I had to pick up a whole different set of skills (Python, R, Spark). I was excited and up for the challenge. I can never give enough credit to Code Academy! Their python course is absolutely brilliant and the best thing about it, it is free. After completing their python course in a week or two, I knew enough now to tackle some data science tutorials. I can’t give enough credit to all the contributors on Kaggle and various data science blogs and platforms. My covert operation involved coding (on the tube to and from work), reading blogs about Data Science and getting myself familiar again with advanced statistics.We truly live in an open source age. After completing a few tutorials online, I was pretty certain this was the right trajectory for my career. THE JOURNEY! At this stage I had looked online on job boards and had been discouraged by the myth that all Data Scientists had to have a PhD or MSc at the bare minimum. I nearly gave up on the dream because I didn’t want to study for another 2 to 6 years before having any slight chance of being looked at for a job offer as a Data Scientist. I was getting frustrated, I wanted to change career but I didn’t have the complete skill set or experience needed to make the switch. Not giving up on the dream, I decide to carry on with my self-study and purchased an Udemy course on Python & R Machine Learning. This was one of the best decisions I ever made. The course cost the price of a Nando’s meal and it give me good foundational knowledge in Classification, Regression & Deep Learning models. I started my job application process and was pretty much knocking on any door that was looking for a Data Scientist. I had completed two projects at this stage and was quite surprised when I started getting interviews. THE OFFER! Each rejection I received was a learning curve for me. I always made sure I asked for feedback to know what to improve on next. Anyway it wasn’t long before someone took a chance on me and got my offer as a Data Scientist. 3 years later, I have been immersed in the world of Data Science for a while, I am passionate about coaching and teaching people how to make that transition from a developer or data analyst to a data scientist. Follow my journey on Instagram below! Fiifi || Data Sciencist (@data_coach) * Instagram photos and videos 144 Followers, 56 Following, 38 Posts - See Instagram photos and videos from Fiifi || Data Sciencist (@data_coach)www.instagram.com

Data Analyst to Data Scientist

3

data-analyst-to-data-scientist-1232da18b07c

2018-04-02

2018-04-02 14:44:41

https://medium.com/s/story/data-analyst-to-data-scientist-1232da18b07c

false

806

null

Data Science

data-science

Data Science

33,617

Fiifi Arthur

null

e19fd25ab257

fiifi.aarthur

2

20,181,104

null

0

null

0

961abc9ded16

2017-12-13

2017-12-13 02:09:32

2017-12-13

2017-12-13 17:25:37

1

true

en

2017-12-13

2017-12-13 17:25:37

15

12331bef8c63

5.075472

92

0

Insights from Coinbase co-founder Fred Ehrsam

5

An Insider’s View of Bitcoin, Blockchain, and the Future of Money Insights from Coinbase co-founder Fred Ehrsam Just another year of up and to the right. I host a podcast which takes deep dives into science, tech, and sociological topics. I do this via interviews with world-class experts who have the patience to engage in truly unhurried discussions of their fields. My episodes are untethered from the headlines, as they’re meant to resonate with future as well as present-day listeners, ideally over a span of years. Occasionally, though, things happen to line up with current events. And was that ever the case this week — as today’s episode is about cryptocurrencies, which have been repeatedly dashing price records, even as Bitcoin debuted on the futures markets this past Sunday. My guest is Coinbase co-founder Fred Ehrsam. With over 13 million users, Coinbase stores more cryptocurrency than anyone else, and the total monthly volume of trades between currencies on its platform stretches into the double-digit billions. You can find our interview right here — or you can hear it on your smartphone by typing “After On Podcast” into the search field of your podcast app. Before we sat down, Fred and I put significant thought into structuring our conversation as both a rigorous introduction to cryptocurrencies (the first half hour or so), as well as an advanced discussion of the more wild and speculative things which could emerge from all this. Our prep, the interview itself, and the research I did in connection with it was a great education, and I’d like to share some of the more intriguing things that emerged from it. Fred discusses the financial incentives which radiate through crypto ecosystems in more nuanced ways than I’d previously encountered. Those enjoyed by cryptocurrency miners, of course, are obvious. In exchange for running the servers that secure the networks, maintain their ledgers, and process transactions, miners get transaction fees from users, and also collect the new coins which enter money supplies at carefully-metered rates (if you’re new to this, Fred and I discuss miners in detail at timestamp 21:10 in the interview). What I’d never considered are the massive incentives facing the developers of open cryptocurrency software. The social and psychic rewards of traditional open source are certainly there for them, along with a certain communitarian ethic. But before Fred mentioned it, I hadn’t considered that most of these folks are also deeply bought into the currencies they work so hard on. In retrospect, this seems head-smackingly obvious. If you pour your evenings and weekends into building something, of course you’re going to use the thing you’re building. And if the thing you’re building is money, you’ll inevitably buy at least some of the currency you’re creating. You hopefully won’t go into debt betting on it (please don’t!). But if you were an early developer on Bitcoin, Ethereum, or another successful project, a minor bet would be enough to make you hugely vested in its ongoing success today. The diffuse rewards of open source have long created amazing things without this financial dimension. Linux and Wikipedia should rank with the Great Pyramids in the pantheon of human wonders (and as a former Cairo resident, I don’t say this lightly). Now couple that power with hard economics — itself the world’s mightiest organizing force — and you really have something. While there are many reasons why BitCoin is worth $300 billion, a giant one is that it’s the super-sexy open source project that paid. Fred astounded me by saying that though hundreds have made real contributions to the codebase, BitCoin’s truly core developers number perhaps fifteen. Brilliant newcomers would no doubt be welcome to join their ranks and make major contributions. But that would mean joining a project whose old-timers have all banked fortunes, which your own voluntary efforts could never hope to replicate. This surely helps explain the explosion in new currencies. Because unlike the situation in other open source areas, being established and well-known will reduce, not enhance, a project’s allure to developers in crypto. Much as I love the ethos of Mozilla and Burning Man, this doesn’t bother me, because I also love Google and the Rolling Stones. Magnificent things — even great art — can be made by folks who are deeply interested in making money. So by extending incentives in subtle new ways, the blockchain and crypto might just create emergent wonders, which the world would not otherwise see. Tech has always aligned incentives more deliberately and aggressively than any other industry, to its great benefit. Facebook achieves things that Clorox and Ford can’t dream of in part because it can make its employees stupidly rich. And here, it’s worth noting that when a cryptocurrency really succeeds, the upside radiates beyond its developers to include much of its user base (as anyone who got into BitCoin before, say, last week will attest. For now). And what if Facebook not only cut its employees in on the deal, but its users as well? Of course we can’t know, because corporate structures don’t support such things. Fred — whose perspective on this extends clear back to the Dutch East India Company (timestamp 53:37), and indeed far beyond that — enjoys speculating on how blockchain incentives and governance might enable entirely new methods of organization. We discuss the notion of an Uber-like entity that’s blockchain-based — perhaps collectively owned by its drivers, or even its users (timestamp 1:03:02), and indeed at least one entity is chasing this very dream. One of Fred’s more fascinating topics is blockchain-based AI (timestamp 1:21:54). The precept is that AI’s lifeblood is great masses of data, for training and honing its output. And however much data Google has, the rest of us combined surely have more. A blockchain-based AI, which takes data contributions from one and all, and rewards data donors when their contributions are used, might therefore become far mightier than any proprietary AI. I find this notion riveting. But a huge countervailing force will be the vast relative efficiency of proprietary networks over decentralized ones that securely leverage the contributions of unknown (and therefore untrusted) contributors. Bitcoin’s founding genius lies in enabling the latter type of network, by cracking the famous “Byzantine general’s problem” (timestamp 23:30). But this comes at great cost, in terms of the compute cycles demanded by its underlying cryptography and “proof of work” requirements. How great a cost? Well, the carbon footprint of Bitcoin’s infrastructure is said to rival that of Denmark. And we can safely assume the Visa card network needs far less juice than six million Europeans in chilly latitudes. Yet Visa can handle 47,000 transactions per second, vs. a single-digit number for Bitcoin. And as AI is orders of magnitude more computationally demanding that ledger transactions, an open, blockchain-based AI will have to overcome tremendous headwinds. That said, Fred’s points about the immense rallying power of financial incentives are spot-on, and armies of brilliant programmers are just starting to explore the blockchain’s potential as a computing platform. Perhaps in the end it will be a horserace between the power of incentives vs. the efficiencies of centralized networks? If these subjects interest you, you should enjoy our interview — particularly if you’re seeking a rigorous primer on cryptocurrency basics, which we start out with (experts may want to jump to the 30-minute mark, or thereabouts). Topics covered in prior episodes of the podcast include neuroscience and consciousness, drones, augmented reality, Medium itself (an interview with Ev Williams), Fermi’s Paradox, and more. I’ll add that my guests have included Chris Anderson (who runs the TED organization), legendary tech observer and participant Tim O’Reilly, and the ever-controversial Sam Harris.

An Insider’s View of Bitcoin, Blockchain, and the Future of Money

530

an-insiders-view-of-bitcoin-blockchain-and-the-future-of-money-12331bef8c63

2018-04-18

2018-04-18 08:28:45

https://medium.com/s/story/an-insiders-view-of-bitcoin-blockchain-and-the-future-of-money-12331bef8c63

false

1,292

Covering the biggest shift in business and society since the industrial revolution

null

newcofestivals

null

NewCo Shift

editorial@newco.co

newco

BUSINESS,WORK CULTURE,STARTUP,ENTREPRENEURSHIP,CAPITALISM

newco

Bitcoin

bitcoin

Bitcoin

141,486

Rob Reid

Podcast host at after-on.com Author (“After On,” “Year Zero,” etc). Founder, of Listen (which created the Rhapsody music service). Tech investor. TED Talk-er.

f849b30c4ddd

RobReid

4,473

168

20,181,104

null

0

null

0

a18ed49f42e7

2017-10-12

2017-10-12 21:21:10

2017-10-12

2017-10-12 22:07:47

6

false

en

2018-01-10

2018-01-10 20:46:41

11

1234d735d6c9

7.931132

25

0

Why Interpretability Matters in People Analytics

5

Interpreting Machine Learning Models Source: Antonio Robers on Flickr Visit us at https://www.ansaro.ai to learn more about how we’re using data science to improve hiring Why Interpretability Matters In the mid 1990s, a national effort was undertaken to build algorithms to predict which pneumonia patients should be admitted to hospitals and which treated as outpatients. Initial findings indicated neural nets were far more accurate than classical statistical methods. Doctors, however, wanted to understand the “thinking” behind this algorithm, so statisticians catalogued “decision rules” from the more easily interpreted regression results. It turned out that both the regression and the neural net had inferred that pneumonia patients with asthma have a lower risk of dying, and shouldn’t be admitted. Obviously, this is counterintuitive. But it reflected a real pattern in the training data — asthma patients with pneumonia usually were admitted not only to the hospital but directly to the ICU, treated aggressively, and survived. [1] Had this “high-performing” neural net been deployed in a clinical setting, it could have caused unnecessary deaths. Only by interpreting the model was a crucial problem discovered and avoided. Understanding why a model makes a prediction can literally be an issue of life and death. As algorithms are used to make decisions in more facets of everyday life, it’s important for data scientists to train them thoughtfully to ensure the models make decisions for the right reasons. Many machine learning textbooks present students with a chart that shows a tradeoff between model interpretability and model accuracy. This is a heuristic, but many students come away thinking that this tradeoff is as strict as a law of physics. In this post, we’ll explore (and question) this tradeoff, offer a framework for measuring interpretability, and apply that framework to a few common models. First, let’s return to why interpretability matters — and when it doesn’t. As the pneumonia example illustrates, interpretability is key for “debugging” models. It’s required in regulated industries like finance and healthcare to audit the decision process and ensure it’s not discriminatory. The US Fair Credit Reporting Act requires that agencies disclose “all of the key factors that adversely affected the credit score of the consumer in the model used, the total number of which shall not exceed four” — and they’re not wrong to set this limit. Models implemented in popular software packages can easily accept thousands of data points, and a huge feature set can quickly make a straightforward explanation nearly impossible (not to mention that collinear features can complicate things further, but we won’t address that here). Interpretability is also key to winning trust in algorithms that try to improve upon human judgement, instead of just automating it. Take our work at Ansaro trying to predict which job applicants will perform best. Our goal is to do better than human intuition. For users to accept our predictions, they have to understand them. So when can we deprioritize interpretability? We think these criteria are reasonable guidelines: Global Interpretability: How well can we understand the relationship between each feature and the predicted value at a global level — for our entire observation set. Can we understand both the magnitude and direction of the impact of each feature on the predicted value? Local Interpretability: How well can we understand the relationship between each feature and the predicted value at a local level — for a specific observation. Feature Selection: Does the model help us focus on only the features that matter? Can it zero out the features that are just “noise”? It’s important to note we’re not talking about interpreting model accuracy — we assume that models have been cross-validated using train, validation, and test datasets, and that an appropriate evaluation metric like AUC or F1-score has been chosen. We also assume that the feature set has been chosen thoughtfully, though this is a big assumption — and one with many interpretations. Next, we’ll apply this framework to a few common model types, to get an idea of how strict the accuracy–interpretability tradeoff really is. Linear Regression We’ll start with linear regression. There’s a reason that linear regression has been the go-to model for the scientific community for the past century — because it’s the gold standard in interpretability. An ordinary least squares (OLS) model generates coefficients for each feature. These coefficients are signed, allowing us to describe both the magnitude and direction of each feature at the global level. For local interpretability, we need only multiply the coefficient vector by a specific feature vector to see the predicted value, and the contribution of each feature to that prediction. A classic OLS regression doesn’t eliminate noise features, but we can accomplish that by removing features for which the confidence interval crosses zero and rerunning the model. Or we can use slightly more sophisticated methods, like Ridge or Lasso regression, that essentially zero out noise features. Random Forest In the middle of the accuracy-interpretability spectrum are random forests. We’ve often seen them described as “black boxes,” which we think this is unfair — maybe “gray” but certainly not “black”! Random forests are collections of decision trees, like the one drawn below. The splits in each tree are chosen from random subsets of our features, so the trees all look slightly different. A single tree can be easily interpreted, assuming it is not grown too deep. But how we can interpret a random forest that contains hundreds or thousands of trees? Many implementations of random forest classifiers include out-of-the-box methods for quantifying the overall magnitude of each feature. For example, scikit-learn’s RandomForestClassifier.feature_importances_ allows us to assess the relative importance of features with one line of code. Feature importances, when used with proper cross-validation, can also allow us to identify the features that are pure noise. However, understanding features’ directionality is more difficult. We can quickly identify that Feature X may be the most important, but does it make Outcome Y more or less likely? There may not be a yes-or-no answer. Unlike a linear regression, random forests can identify non-monotonic relationships (a big part of the reason they outrank regression on the accuracy axis). In one region of the observation space, a feature’s direction may be positive; in another it may be negative. Understanding how a particular observation’s features contribute to the prediction is also challenging, but doable. To achieve local interpretability, we can catalogue the decision paths for a specific observation through all our decision trees. We then sum the decreases in the Gini index for each feature, across all these paths. For the non-statisticians, Gini index decrease is a measure of how much more “cleanly” the classes are separated after a split in a decision tree. This sounds like heavy lifting, but libraries like Ando Saabas’ excellent treeinterpreter make this practical. [2] Neural Networks As the hottest topic in machine learning over the past decade, we’d be remiss if we didn’t mention neural networks. Hailed for outstanding accuracy in difficult domains like image recognition and language translation, they’ve also generated criticism for lacking interpretability: “Nobody understands how these systems — neural networks modeled on the human brain — produce their results. Computer scientists “train” each one by feeding it data, and it gradually learns. But once a neural net is working well, it’s a black box. Ask its creator how it achieves a certain result and you’ll likely get a shrug.” — Wired Magazine, October 2010 [3] We think that’s a dramatic overstatement. We’re also cognizant of the fact that there are many types of neural network architectures, and making blanket statements about them is difficult [see the Asimov Institute’s terrific Neural Net Zoo]. For simplicity, we’ll focus on convolutional neural nets (CNNs), widely used for image recognition (and many other applications). Imagine we’re training a CNN to predict the probability a 64x64 pixel image contains a cat. We could start to gain insight into how important features are by tweaking them and seeing how the resulting probability changes. But the features fed into this CNN are 4,096 RGB pixel values. Knowing that a particular pixel corresponds with being a cat isn’t particularly useful, nor would we even expect such a relationship to occur — pixels representing a cat could appear anywhere in the image. For this model to be semantically interpretable, we need to understand its features at a more abstract level. However, building up raw pixel values (or any high-dimensional data like audio waveforms or unstructured text) into abstract features through the network layers is where interpretability breaks down, where we start to lose the global understanding of what a specific feature contributes. The ability to encode non-linear relationships across such an array of features is where neural nets outperform many other models in terms of accuracy. Despite this complication, all is not lost in terms of interpretability. Back to our example. Instead of tweaking individual pixels, we can track and inspect the training images that maximally activate neurons. By looking at neurons at deeper levels of the CNN, we may be able to find neurons that correspond to semantically meaningful concepts like “ear” or “tail.” We can then track the weights assigned to the neurons that we believe represent abstract concepts by the network’s final layer, giving us a rough idea of global feature importance. For local interpretability, we can use occlusion to understand where a CNN is “paying attention.” We iterate across the image, setting a patch of pixels to be zero, running the occlude image through the CNN, and logging the “cat” probability. We can then visualize the contribution of each part of the image to the “cat” probability as a 2D heat map. These methods aren’t as simple as examining coefficients, but it shows that neural networks are not completely black boxes. Concluding Thoughts Rather than being a static tradeoff, we think of accuracy-vs-interpretability as a frontier, one which is constantly being pushed outwards. Over the next decade, we believe developing more interpretable models will be as important as developing more accurate models for the data science community. Much of the work on advancing interpretability will be done by domain-specific experts — better ways to visualize CNN results, for example. But there are also exciting advancements in approaches that transcend specific model types. One approach that’s caught our attention is LIME (Local Interpretable Model-Agnostic Explanations) [4], which allows you to build any model you like, then use perturbation and linear approximation to explain specific predictions. As machine learning becomes more important in our day-to-day lives, promoting trust in good algorithms — and the ability to detect bad algorithms — is critical. So, too, is designing algorithms from the start with interpretability in mind. –Matt & Sam, Cofounders, https://www.ansaro.ai matt.mollison@ansaro.ai, sam.stone@ansaro.ai We’re hiring! If you’d like to help companies make the best hiring decisions, check out our job postings: https://angel.co/ansaro/jobs References [1] “Intelligible Models for HealthCare: Predicting Pneumonia Risk and Hospital 30-day Readmission,” Caruana, et al. (2015). http://people.dbmi.columbia.edu/noemie/papers/15kdd.pdf [2] http://blog.datadive.net/interpreting-random-forests/ [3] https://www.wired.com/2016/10/understanding-artificial-intelligence-decisions/ [4] http://blog.fastforwardlabs.com/2017/09/01/LIME-for-couples.html

Interpreting Machine Learning Models

212

interpreting-machine-learning-models-1234d735d6c9

2018-06-10

2018-06-10 12:19:45

https://medium.com/s/story/interpreting-machine-learning-models-1234d735d6c9

false

1,850

We're data scientists, HR professionals, programmers, psychologists, and business analysts. We're united by our love for hard problems and our belief that data science can make work better. Our website: https://www.ansaro.com

null

Ansaro Blog

hi@ansaro.ai

ansaro-blog

DATA SCIENCE,PEOPLE ANALYTICS,HUMAN RESOURCES,MACHINE LEARNING,HIRING

ansarotech

Machine Learning

machine-learning

Machine Learning

51,320

Ansaro

null

7eb91541c14c

Ansaro

20

1

20,181,104

null

0

null

0

f702855ffe47

2017-11-14

2017-11-14 00:00:26

2017-11-14

2017-11-14 00:00:27

9

false

en

2017-11-14

2017-11-14 00:00:27

12

1234dd072be0

2.6

0

null

3

How to Robot Proof your Education and Career # medium.com We sometimes like to joke about the arrival of “robot overlords”, however, at the rate in which AI and robot… This Digital Giant Outperforms through Data and AI # medium.com The last few years we have seen a solid stock market growth in tech companies. However, one of the giants re… The Dangerous Problem Within Restaurant Marketplace Online Ordering # medium.com Why Many Forward-Looking Restaurants Actually Losing Money With Each Online Order They Receive Via Marketpla… Facebook tackles… revenge porn? # medium.com Facebook are trialling another of their out-of-the-box ideas to combat social issues. Following my write-up … Perception Models for Self-Driving Cars with Jianxiong Xiao # medium.com TWiML Talk 058 We are back with our second show this week, episode 2 of our Autonomous Vehicles Series. This… Training Data for Autonomous Vehicles with Daryn Nakhuda # medium.com TWiML Talk 057 The episode you are about to hear is the first of a new series of shows on Autonomous Vehicle… SIIA, IEEE, and AI Caucus Event on Machine Learning and Ethics # medium.com This first appeared in the SIIA Digital Discourse Blog. On November 7, 2017 I made a short presentation to t… Human Factors in Machine Intelligence with James Guszcza # medium.com TWiML Talk 056 As you all know, a few weeks ago, I spent some time in SF at the Artificial Intelligence Conf… As you all know, a few weeks ago we spent some time in San Francisco at The AI Conference. # medium.com As you all know, a few weeks ago we spent some time in San Francisco at The AI Conference. While there, I sa… The Matrix revisited: Part I # bernadette.life Copyright Warner Bros In 1999, my mother suddenly had an urge to go to the cinema. She took me and a friend … 5 professions that could see significant growth with the rise of AI # venturebeat.com GUEST: The words “artificial intelligence” often conjure up a sense of fear and apprehension. Fear for the u… AI’s stock trading potential is probably overhyped # venturebeat.com GUEST: Automated trading algorithms fueled by artificial intelligence have quickly taken over the financial …

12 new things to read in AI

0

12-new-things-to-read-in-ai-1234dd072be0

2017-11-14

2017-11-14 00:00:29

https://medium.com/s/story/12-new-things-to-read-in-ai-1234dd072be0

false

371

AI Developments around and worlds

null

AI Hawk

aihawk1089@gmail.com

ai-hawk

DEEP LEARNING,ARTIFICIAL INTELLIGENCE,MACHINE LEARNING

null

Deep Learning

deep-learning

Deep Learning

12,189

AI Hawk

null

a9a7e4d2b403

aihawk1089

15

6

20,181,104

null

0

null

0

null

2018-07-03

2018-07-03 08:13:57

2018-07-03

2018-07-03 08:20:48

3

false

en

2018-07-03

2018-07-03 08:20:48

3

1236043b4a85

2.806604

0

Data-driven organizations are 23 times more likely to acquire customers, six times as likely to retain those customers, and 19 times as…

5

How to develop data-driven organisation? Data-driven organizations are 23 times more likely to acquire customers, six times as likely to retain those customers, and 19 times as likely to be profitable as a result. Taking decisions based on Data not only makes instinctive sense, but the evidence is mounting that it makes strong commercial sense too! For instance, the McKinsey Global Institute indicates that data-driven organizations are 23 times more likely to acquire customers, six times as likely to retain those customers, and 19 times as likely to be profitable as a result.While awareness of this potential is undoubtedly valuable, this knowledge needs to be converted into actions. So how do you go about becoming a data-driven organization? Everyone has heard about the power of data, but every few know how to utilize data to transform and gain competitive business advantage. This article is an attempt to lay down 10 practical steps for building a data-driven organization. 1. Ask questions: which of your most pertinent business problems are likely to be solved using data. 2. Organise a workshop or send your leadership team to some analytics, data science conference to gain insight on how other organisations are transforming using data. Explore Data Science Congress: Let your leaders learn from world leaders in Analytics, Data Science, Predictive Analytics, BI, Cognitive etc, how they have transformed Google, IBM, Vodafone, AIG, Airtel, State bank of India, Oracle, Microsoft, Cisco, Flipkart etc Learn More 3. Build a team: get the best talent in data science and analytics on board. Hire IBM Certified Data Science Talent: Get in touch with Career Management team at Aegis to hire IBM certified best brains in Data Science, Business Analytics, BI, Big Data, and Machine Learning etc View Online Placement 4. Develop skills and competency for developing data driven organization: Your employees are the best resource as they know your business problems as also the opportunities you are missing. Train them on new technologies and tools like Watson Analytics, Predictive analytics. Aegis and IBM launched Post Graduate Program in Data Science, Business Analytics and Big Data to develop high-end skill pool in India. You can reach out to Aegis Executive Education and Career Management Team to design & develop a road-map for achieving this objective and of course hiring the ready pool of zero to 30 years’ experience analytics professionals. View Program 5. Get the relevant data which can provide deeper insight to solve pre-defined problems. Your team might come across some missing business opportunities while analyzing such data. 6. Build Data infrastructure: As we did for one of the leading Internet Service provider in Mumbai using Hadoop and Spark. 7. Empower your leadership team and stakeholders with data visualization tools: Business Intelligence tools like Tabelau, Qlik, Cognos or open source tools like R, Python. Nominate your employees for training on these tools in Data Science Congress on 5 June in Mumbai. 8. If you have already adopted some or all of the above, move to next level: Develop sophisticated predictive models. Train your people on machine learning, predictive modeling and statistical techniques using open source tools like R, and Python; or deploy ready to use proprietary platforms like SPSS, Watson Analytics, SAS etc 9. Once you record successes from your initiatives, try percolating the data-driven culture across the organization by publicizing these success stories. Train and motivate more employees. Please do share your story with us. 10. Keep implementing data-driven processes and tools to automate the routine task for better productivity, efficiency, and enhanced customer satisfaction.

How to develop data-driven organisation?

0

how-to-develop-data-driven-organisation-1236043b4a85

2018-07-03

2018-07-03 08:20:48

https://medium.com/s/story/how-to-develop-data-driven-organisation-1236043b4a85

false

598

null

Data Science

data-science

Data Science

33,617

Bhupesh Daheria

null

56cf668ef1cc

bd_70237

1

20,181,104

null

0

null

0

null

2018-08-12

2018-08-12 01:03:14

2018-08-12

2018-08-12 10:37:52

8

false

en

2018-08-12

2018-08-12 12:31:47

1

123693e36ba9

7.857862

7

0

Every year summer melts the locals and tourists in Manly Corso into a thick paste of thirsty beachgoers. As you sludge away from the beach…

5

Hijacking A Thief’s Mind Rooms By The Sea — Edward Hopper Every year summer melts the locals and tourists in Manly Corso into a thick paste of thirsty beachgoers. As you sludge away from the beach and to the crowded vendors, its consistency is unremarkable — a blur of unforgiving pedestrian-traffic and a creeping haze of sweat and vaporising sunscreen. As repetition begins to define itself at the beginning of November, the workers that hold up the sides of Manly Corso descend into homogeneity. A cycle of demand fastens till it reaches constant velocity. As the arena for predictability rises, the workers let down their own guards as they take their seats for the Summer rush. No one wants to stand and get in someone’s way. Which is what a thief wants: consistency. “‘Be yourself’ is about the worst advice you can give some people” — Thomas Lansing Masson By December she knew when I finished my shift. She knew where my locker was. By January I knew she had my phone. Yesterday in Seattle, an aircrew ground worker stole a turboprop propellor plane, flew for an hour and then fatally crashed into a wooded island. It was the real-life Grand Theft Auto. The pilot even said it himself: Air-Traffic Controller: Right now he’s just flying around, and he just needs some help controlling his aircraft. Pilot: Nah, I mean, I don’t need that much help. I’ve played some video games before. When the plane’s fuel tank began to empty at 7,500 feet, the pilot opened up: I’ve got a lot of people that care about me. It’s going to disappoint them to hear I did this. I would like to apologise to each and every one of them. Just a broken guy, got a few screws loose I guess. Never really knew it, until now. You can tell a plane has crashed with a full-tank of fuel by the fireball that follows. As the kerosene burns around the passengers, their clothes and watches melt into their skin. For investigators, the wreckage is a paste of burnt steel and flesh. Amongst this mess, reasons and answers need to be found. But when a plane crashes by running out of fuel, the reason splinters out from the ground. Even without the scorched earth, there’s still a scorched mind somewhere in the crash. Some loose screws let the furnace explode. “Clean the air! Clean the sky! Wash the wind! Take the stone from the stone, take the skin from the arm, take the muscle from the bone, and wash them. Wash the stone, wash the bone, wash the brain, wash the soul, wash them wash them!” — T. S. Eliot, Murder in the Cathedral Why did she steal my phone? Why did he steal the plane? At this moment it’s easy to connect different characteristics and underline them as “motivations”. It wouldn’t be uncommon to hear the following: “She had a broken childhood” “His work was too much pressure” “Her friends all do it” “He was mentally ill” The majority of these comments are environmental influences. They linger in peoples’ minds, edging and scratching away at new ideas until they burst into our conscious as ‘our’ choice. But when someone’s described as ‘mentally ill’, we introduce a seismic shift into how we can ‘be’. It’s no longer ideas fighting for attention as we pass by new scenery. In this case there’s no rules — it’s a brawl in the dark. No one knows who’s fighting who. As the silhouettes grapple with each other, it’s only the ones that fall into stillness that tell us what’s lost. There’s nothing more biologically fascinating than the human brain. As we dig our way through craniums, we find the teeth marks of evolution having torn into us again, and again. The electric paste that now floats in each of our heads is like bubble-wrapped popping candy. Billions of muffled explosions showering the mind with sugar to keep it alive. Yet quiet enough that you can forget how bloody it once was. Think about how many minds have been lost to experimentation to get you here The four lobes of the brain are the Frontal (working-memory, planning, motivation), Parietal (sensory information), Occipital (vision processing) and the Temporal (derives meaning from sensory information). Neural tissue can be further divided into either grey or white matter. Grey tissue is used for information processing, where electric impulses converge into actionable opportunities. To facilitate these impulses, white matter helps propagate electric energy through the type of neurons found within. At the light microscope level (1 millimetre — 1 micrometre) we can distinguish the cytoarchitecture of nerve cells. With their soma (cell bodies) ranging in diametre from 10–50 micrometres, they can be up to twice the width of red-blood cells. The distinction between nerve cells and the rest that make up our bodies, is their ability to innervate others. Like a chain reaction in a nuclear bomb, the human brain is an infinitely recursive set of reactions being ignited from the world around you. In thanks to the pitch perfect harmony between your motor and sensory systems, you’re able to react to this world and live for another day. ‘Living’ is all evolution wants. It doesn’t care about a species’ integrity — it cares about whether there’ll be a tomorrow. Such persistence across time has allowed phylogeneticists to map the evolution of neurons from simple nerve nets to the complex domains they inherit today in the central and periphery nervous systems. When eukaroytes (plants, animal and fungi) split from the prokaroytes (bacteria) more than a billion years ago, multicellular organisms formed and separated into the five major animal groups of today: Ctenophora, Poriphera, Placozoa, Cnidaria and the more than two dozen phyla that make up Bilateria which includes chordates such as humans. From this split, brain development has uniquely emerged at least in five separate radiations of vertebrates. The kinetic energy of cephalisation that concentrated neural activity into human heads ensured that the Homo gene was delivered a decisive strategic advantage within the animal kingdom. It’s not bragging if you can back it up. But what happened within our brains that made us the dominant species? There’s approximately 100 billion neurons in a human brain but they’re outnumbered 10 : 1 by Glial cells. Neuronal function isn’t a binary matter. It’s as rich as dining experiences that are engineered by a near infinite list of parametres: the menu, the weather, the waiter, the entreé and so on. A neuron identified through Gogli staining Neurons are excited by a changing chemical gradient around their neuronal membrane. As ions permeate through sodium, calcium and potassium channels, the changing electric difference between the extracellular fluid and the cytoplasmic substance leads to a rise from the resting potential of -65 millivolts until it crosses threshold at approximately -30millivolts. At this moment, an electric signal charges from the axon hillock to the axon terminals within 1000th of a second. When this action potential reaches the axon terminals, a neurotransmitter is released from the presynaptic neuron so it can communicate to the post-synaptic neuron by connecting with its dendrites. The neurotransmitter acts as an effector molecule, enabling the propagation of the action potential across the nervous system. But what’s to guide these neurotransmitters from the presynaptic to postsynaptic neuron? Thanks to the structural support of Glial cells, the brain has limited free space. Astrocytes invest synaptic clefts to maintain the brain’s biochemistry so there’s no concentration of any particular neurotransmitter. How about the speed of these action potentials? Why do they move so fast in the first place? In the Central Nervous System, oligodendroglial cells envelop the axons of neurons to create an internode/Node of Ranvier pattern called myelination that not only improves the conducting speed of the impulse, but also reduces the leakage of ions into extracellular fluid. Similarly, in the Peripheral Nervous System, Schwann Cells myelinate the peripheral nerves that enable us to react and sense our environments. As opposed to the simple network of neurons that are usually heralded as the architects of human behaviour, there’s instead an overwhelming grasp on the brain that’s controlled by glial cells. Neurons wouldn’t be able to fire their action potentials if astroglial cells couldn’t absorb glutamate from the synaptic cleft and then create glutamine from ammonia (via glutamate synthetise), because glutamate is an excitatory neurotransmitter. Unfortunately, excess glutamate behaves as a excitotoxin that leads to neuronal damage. Instead of the brain being just 100 billion neurons, it’s 1 trillion glial cells and 100 billion neurons. Together, these networks have made us human. Even when we don’t know what’s going on inside someone’s head, we do know what’s there. On the course to machine intelligence there’s a branch approaching that leads to Whole Brain Emulation (WBE). It sounds as it reads. By uploading someone’s mind to a computer, it’ll be possible to model somebody’s behaviour and the derivates that come with it. For instance, prediction. Successful WBE requires an innate understanding of neural and cytoarchitecture that’s beyond our levels today. Only in the last decade or so are glial cells receiving the attention they deserve, previously thought to be just “space-fillers” in the brain. While WBE may be used to understand the minds of those that escape our logic today, they may lead to inescapable realities. For anyone that misbehaves, it’s easy to discern their behaviour as a byproduct of their environment. That’s why we hear… “She had a broken childhood” “His work was too much pressure” …after any malicious, confusing or simply inhumane act. But what if within the network of neurons and glial cells, there was a menacing set of wires that led to someone’s mind becoming entangled in a toxic worldview that strangled any sense and empathy from them. When gametes fuse together during sexual reproduction, zinc atoms burst from the egg marking a successful fertilisation. It’s easy to forget that in this moment, two individual cells have mixed together and are about to form a zygote. In that moment, all the genetic material needed to form a complete human has been processed. Life has begun. The mind floods open. Remember, “that” person came from just one cell. Classification of proteins that are found in the human genome’s 3,088,286,401 letters of DNA The comprehensive goal of computing is to emulate and improve upon, human cognition. Presently, computers demonstrate discrete abilities that are mostly inspired by either human abilities or lack of. WBE represents the pivotal moment when computers are able to emulate human brains. A technical achievement that will be perhaps one of humanity’s greatest. At long last, the inner workings of someone’s mind will finally be able to be cracked open and the paste of ‘us’ can be filtered. Each constituent, dissected and categorised so the precision of the human brain can be unveiled. For programmers that are currently working directly or indirectly towards WBE through machine-learning research and neuroscience, it’s not enough to consider the brain as a network of neurons. So why are all machine-learning algorithms obsessed about neural networks and not the glial networks that outnumber them 10 : 1? Cheers for reading, hope you’ve enjoyed it! Bryan Jordan - CEO, Co-Founder - Bright | LinkedIn View Bryan Jordan's profile on LinkedIn, the world's largest professional community. Bryan has 4 jobs listed on their…www.linkedin.com

Hijacking A Thief’s Mind

17

hijacking-a-thiefs-mind-123693e36ba9

2018-08-13

2018-08-13 11:22:06

https://medium.com/s/story/hijacking-a-thiefs-mind-123693e36ba9

false

1,782

null

Neuroscience

neuroscience

Neuroscience

6,742

Bryan Jordan

Electrical engineering/Neuroscience student at University of Sydney.

aa1d1923f016

bryanjordan

61

393

20,181,104

null

0

null

0

null

2017-11-25

2017-11-25 03:34:23

2017-11-25

2017-11-25 03:35:53

1

false

en

2017-11-25

2017-11-25 03:35:53

0

12378da077eb

1.901887

0

You wake up from a long sleep and it’s 2050 and much anticipated future is finally here. We would be working under robotic boss who keeps a…

5

Glance of Tomorrow You wake up from a long sleep and it’s 2050 and much anticipated future is finally here. We would be working under robotic boss who keeps a track record of employees health statistics ,So I can’t take a fake health leave. My office’s automated airbus would reach my home with my breakfast being served inside just to save my time and earn an extra revenue for the company . We would be flying to office in a airbus causing air traffic everywhere. We would desperately wait for a vacation which our calender couldn’t figure out.Earth would cause boredom because we have explored every corner.Our bucket list would comprises of human free location.New planets will be our new holiday destination. Finding more human species on other planets than on earth wouldn’t be startling. Setting reminder to communicate with people will be on our to-do list target, Our time would get inadequate and leisure would be jammed with tension. Everyone would have a personal assistant who will be smarter than themselves. Our personal assistants would be a clone of ourselves with higher intelligence and common sense. Multitasking would no longer be a cumbersome process. No more limits for new dimensions. Space scientist would have explored the universe and finally return to earth to research on it’s apocalypse.We would wake up everyday in a newer dimensions but being haunted by the past dimensions. Extra terrestrial species will become our friends and people around us will become strangers. It will be the anomaly of the future. Smart-phones would become smartest-phones, which can track your brain waves for inputs without physical interaction.Burgers and pizzas ordered will be delivered at the speed of light but loneliness and depression would be your favorite ingredient. Movies would be boring because actors are replaced by programmable acting-bots better than some present actors. Your pet can understand human language by the translating device he/she wears. Google and Facebook will control you with the data they own about your lifestyle.Privacy will no longer be your concern because you are desperately searching for freedom. They would have disabled the uninstall button from your life.Only choice you are left is to power it off and look the world with your naked eye. Roads will be traffic free but by sky gazing we would find flying vehicle trafficking the air.Pure oxygen would become expensive, covering the import charges and delivery charges. Media coverage would be focused on this issue which would never last more than a day. Hence future would be a destination for latest technology which monopolizes the market by trading human freedom and individuality. But there’s nothing like a beautiful sunset to end an healthy day.

Glance of Tomorrow

0

glance-of-tomorrow-12378da077eb

2018-05-09

2018-05-09 11:20:48

https://medium.com/s/story/glance-of-tomorrow-12378da077eb

false

451

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

karthik ziffer

I’m a Tech Enthusiast , Blogger , Explorer , Web developer , constant learner.

3b97a4e0a437

Karthikziffer

17

488

20,181,104

null

0

null

0

null

2018-05-10

2018-05-10 16:18:07

2018-05-10

2018-05-10 16:21:18

2

false

en

2018-05-10

2018-05-10 16:24:36

1

1237a6577113

11.073899

0

WOW, what a year this has already been. Really making myself known and people from all over the world love these interviews. Something that…

5

An Interview with Alan Grogan- ‘Challenging your Chief data person’ The risk and processes that they will face in 5 years time? WOW, what a year this has already been. Really making myself known and people from all over the world love these interviews. Something that started from nothing has now grown my name to over 20,000 followers and it doesn’t stop there. I remember taking this leap not knowing where the journey will take me but to just keep going on what am doing. So far this journey is a success, and has led me to interview: Michael Qundazzi-At PWC, John White at Inc, Greg Allum- Previous Head of Social Media at Sony, Simon Chan, Team human, Rosa Markarian, previous Vice President Global Product Development,of Xaxis, J.J Delgado previously at Amazon, and soon my interview with the Co-Founder of LinkedIn, Eric Lu will be published soon. And many many more. But it is finally great to interview and get Alan Grogan at ATOS, listed as Top 50 Data Leader, on this list of great people and expand my portfolio to over 15 interviews and growing further. Alan talks, to us about ‘Challenging your Chief data person’ The risk and processes that they will face in 5 years time? One that will appeal to encapsulate the brightest brains to this read and the people that want to learn, what will happen is we are in some sort of transition stage how data will be used, and the trust and knowledge it shares with the company and with the people in hand in charge of it. Data I particularly think is going through a turbulence time on how it is accessed and how it is given. This comes into play also with outside forces, the trust of data and who has what, and the privacy that is starting to circle round, if this is necessary for all data to be recorded, sought or drawn. Alan talks about this from an organisational level and looks upon the highest people in position who are being challenged in this area. Alan explains this perfectly and will draw in many experts who are interested to learn just that. This is an interview not to be missed. I will carry on interviewing the best. Please support my Patreon crowdfunding page to help me to fund these interviews as I do it out of love and I don’t make anything from it- it is you brilliant people like you after all that helps make this community work- Pledge here help me to continue-https://www.patreon.com/NazarethQarbozian 1-Alan, a privilege to interview you. A man who has grown professionally in his career; is a UK Top 50 Data Leader and works for one of the leading Digital Services organisations of the world, Atos, it is an honour to interview you on your thoughts on data and the challenging areas we are going to see. My first question to you is? How is data going to be disrupted in 5 years time when a computer becomes faster and more clever, will the human touch be lost over the machines? Firstly thank you, Nazareth, for giving me the opportunity to be part of your latest series of interviews. I am in a privileged position where I work as the trusted key advisor to a fascinating array of large and generally complex organisations from aerospace to whisky manufacturers. They all want direction and help to realise the benefits of being more data and digitally enabled. Having a really good network and team around me too helps guide my thinking and intuition about key trends, key challenges and where things will end up. I do see intelligent platforms at a level where the human touch is being lost to varying degrees, but all these platforms have been generally designed on a First Principles basis and in a very transparent and collaborative environment. This ensures the end-to-end outputs of the work are clear, so any loss of human touch has been deliberately planned in the design and then tested, piloted, and rolled-out with little or no negative implication. The human touch will be reduced where operation excellence and customer experience requires it, but I just hope that it is counteracted by an increased importance and focus on experiences where human touch still exists. It would be a key advantage for organisations who get this balance right first. I currently believe that the real data disruption will occur behind the scenes over the next five years. We need to understand where the future computing power will come from because to run faster and more intelligent digital systems you need very fast processing. Just last month Microsoft’s CEO warned the world is running out of computing to support future demand and without a concerted effort to move into areas like Quantum Computing, all this energy in the fourth digital revolution may be stifled somewhat. I already see clients starting to evolve their thinking in/ around quantum. Until then, I believe the organisations who drive greater value will be those who have the best hybrid cloud strategies (most large organisations have more than one cloud) to manage their AI, Cognitive and other complex compute-heavy applications. Until we sort the compute situation then we run the risk of stifling the development of intelligent machines being widespread. 2-Challenging your data officer is going to be a significant issue in years to come as organization adapts more to digital transformation and the speed of change that is happening within that area. The Chief Data Scientist has a lot of responsibility for implementation to execution? How will this person be a challenge in the organization? What sort of problems do you think a Chief Data officer will face? The majority of the Chief Data Officer (CDO) role right now is focused on protecting data and preparing it. This is referred to as a ‘defensive’ approach and covers aspects such as data quality, regulation, governance and compliance. Although CEOs want to treat data as an asset, a purely defensive CDO treats data as a liability. We obviously need the defensive elements covered by the CDO but we need to rebalance the overall role to drive value for the organisation, for example, to increase revenue, reduce cost, reduce risk and drive excellence in areas such as Customer Effort, Customer Experience, Customer Excellence and so on. This is easier said than done. How do we get CDOs on the offensive to drive more commercial outcomes for the organisation? Well, just like the best data scientists and data analysts are both technical and business literate and experienced, so are the best CDOs, and just as we have a lack of data scientists, so do we have a lack of CDOs who are essentially experienced entrepreneurial business technologists. The good news is that this pool of value-creating CDOs who take an offensive ‘commercial’ approach whilst covering the defensive aspects of the role is slowly growing. The problems a defensive CDO might find are associated to when ‘the business’ starts to take the lead on the missing and needed value creating aspects. This is where shadow IT arises. If a defensive CDO then pivots to being commercial, there is an issue that the business won’t want to lose control of its investment, learnings and ownership. So the more defensive a CDO is, the greater the risk is of them being disconnected from the customer and business needs. Shifting to the role of Chief Data Scientist, their role has a balance between research and business implementation of worthy solutions. I see a large number of Chief Data Scientists struggle getting their great work delivered into the business and just like the CDO role, they are too much on one side and not balancing this with the right level of direct business sponsorship. The biggest risk they face is being seen as too much as an innovative and exploration team, rather than an actual solution design, agile development and implementation service. Reporting to the CDO, the Chief Data Scientist and their team will need strong leadership and support from the CDO with the right business model that is designed to work closely with the business and deliver their responsibilities in the executive committee signed-off Data Strategy. 3-What likely issue will the Chief Data Officer face when the organization comes in terms with AI? Will they lack skills or knowledge in the way data is transformed and readily implemented or will they have to do things differently to speed up their approach in terms of implementation and how humans and machines will do things differently? Over time analytics has become much easier to implement due to better and cheaper industrial tools. We are now seeing the exact same developments with AI. The same comparison can be drawn with issues AI faces. These issues cover from where there is a lack of a suitable infrastructure on which to build your AI platform right through to the lack of knowledgeable internal resource. Culture will continue to play a massive part in terms of acceptance. Where we’ll see this mostly play out is how people automatically challenge the quality of any data being used as a mechanism to distrust the AI’s quality of output. My work in industry on AI mostly always has had a Data Quality work stream attached to the overall programme and a good CDO will work this to their favour on their Data Management objectives and plans. Hopefully, soon we’ll hear more about how AI is being used to drive data quality as an output of the AI. Regarding any potential skills or knowledge gap, well we know good leaders learn to engage and learn from their staff. So I have a hypothesis that the typically defensive CDO is being impaired from gaining the commercial knowledge by the skills shortage in Data Science and other Digital disciplines, whilst CDOs who grew out of these disciplines have the advantage and they can much more easily learn the risk side of the business. You can say good risk managers are an abundant species, whereas good AI capable data scientists are relatively rare individuals. So we basically have less commercial CDOs because the market is not helping us. The majority of CDOs will need to do things differently through partnerships to speed up their plans. We see such good partnerships in FinTech where AI services companies offer both niche and broad services. But partnering itself also comes with a level of risk, so I’m keen to see what it will take to push more defensive CDOs to take such a risk before their competitors gain the advantage. For a more balanced commercial offensive CDO, collaboration comes far more naturally. They embrace this cross-business cross-party approach with their more entrepreneurial and digitally disruptive mindset. 4-What is going to happen to the Chief Data Person when they are in competition with other businesses to deliver a distribution of information and consultation? Are they in fear that other businesses may have an edge over their implementation of providing solutions? Have data officer roles got even more challenging as IT is moving quite quickly in this area, meaning competition is now everything? Great question. Competition is coming in many ways from both internal and external parties. All too often competition is occurring where generally part of the organisation is not having its needs or wants to be served by the CDO for whatever reason, so they build data capability (people, process, tech, governance, partners, etc.). What follows is not pretty as the CDO grapples to reassert his/ her authority and this becomes self-harming for the organisation as a resource at all levels is tied up in internal politics. If you have the foresight, there are many methodologies and approaches to prevent this whole situation from happening and similarly, tools exist to resolve such unhealthy conflict. Offensive CDOs generally have greater issues than defensive CDOs, which is particularly bad as it’s the offensive CDOs who need less distraction caused by internal politics. Competing with external competition is by far the area of focus that offensive CDOs thrive in. The best ones act like a business within a business. They have their own revenue targets, cost-saving targets, operational KPIs and even sometimes CEX targets. I’ve seen some CDOs use a staff utilisation target, but my view is that this can be a meaningless KPI because every business has data problems and opportunities. If your data team is sitting on their hands then you’ve certainly got some serious leadership, process or cultural issues. So again it’s all about the outcomes the CDO is supporting or responsible for. The edges some CDOs have over their competition come down into 5 buckets — People, Technology, Governance, Process and Data. For example on technology, to support an agile offensive CDO they really need ownership of data and analytics technology and its architecture. I’ve been saying this for years yet it still surprises me that I find too many organisations who do not support this view and generally they have a more defensive CDO at the helm. Why would you not let them own the decision on what tools they can use? On governance, yes you need governance on how they use tools/ technology, how they share data, etc. but it’s not about having the most stringent governance, but it’s about having the right level of governance. There is no competitive advantage in being the most compliant organisation in the world, you just need to be compliant. On your point on whether CDO roles have gotten more challenging with IT, I would say that as the role continues to evolve and organisations get more mature, we are seeing things like AI become more accessible. 5- And my last question to you is they say companies are failing to meet their targets in the implementation of IOT projects (Cisco says this is a worry), and that in the future CEO’s will face even more challenges as they fail to address data transformation in the organisation putting their firm at risk (Forrester, 2018). How is this going to affect the Chief Data Officer when problems like this are going to occur? What will happen? Is the relationship at risk from the top level down? I’m optimistic that as the commercial approach of the CDO becomes more prevalent that we’ll see the CDO being a widely accountable leader and agent for breaking down the barriers that hold back such digital change. This leads to the biggest problem right now in the whole CDO conundrum, which is what is the optimal operating model, including seniority and reporting line, for the CDO. Unfortunately, this question has many answers all down to a sector, budget, company maturity, company culture and even political environment. What I hope for is that the CDO has the seniority to hold all organisational units to account, just like a Finance or HR department works two ways to hold units to account but also to support them. So should an Industry4.0 / IoT programme stall then the CDO, who would have some level of a financial target associated with the project, would have clear escalation through a financial performance review as an example. I still remain positive that the CDO role is maturing at a relatively quick pace so that CEOs won’t have to experience such reported issues when it comes to realising their transformation goals. — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — - Nazareth is a person who is a disruptor. He has interviewed the brightest brains in top-tier companies, such as Sony, Inc, PWC, Atos and LinkedIn and many more. Nazareth grew from 0 to 20,000 followers in 12 months. He has a passion to share real informative content to his readers in real time, and loves to work in the digital scene. He has over 15 interviews to date and it seems it is not stopping there as 2018 is going to be a great year of growth. Nazareth will continue to disrupt and interview the best brains around the world and draw in more growth as people realise the content he is sharing that can prove to be valuable in the tech and digital sectors. If you can help him on his journey please pledge to his crowdfunding page on Patreon. He is trying to inspire and make a movement to help people realise their journey. As little as 3 dollars a month this can be realised. He wants to share his views and carry on inspiring people lives online. https://www.patreon.com/NazarethQarbozian

An Interview with Alan Grogan- ‘Challenging your Chief data person’ The risk and processes that…

0

an-interview-with-alan-grogan-challenging-your-chief-data-person-the-risk-and-processes-that-1237a6577113

2018-05-10

2018-05-10 16:24:37

https://medium.com/s/story/an-interview-with-alan-grogan-challenging-your-chief-data-person-the-risk-and-processes-that-1237a6577113

false

2,833

null

Data Science

data-science

Data Science

33,617

Nazareth Q-Digital Interviewer

b37fccc0461f

nazareth.qarbozian

2

7

20,181,104

null

0

null

0

a605fc00d24c

2018-01-08

2018-01-08 22:18:29

2018-01-08

2018-01-08 22:19:44

2

false

en

2018-01-08

2018-01-08 22:20:52

0

123880ed07b3

2.424843

1

0

As part of my journey on understanding AI this is where I started “What is the difference between AI and ML? I realized by the way that for…

3

What is AI versus ML ? As part of my journey on understanding AI this is where I started “What is the difference between AI and ML? I realized by the way that for many people these are the same thing: / however they are 2 distinct areas of computing, very well connected… So after a bit of research on the internet I took the 2 first link that popped up and here there are, What it AI versus ML? First reference on google Quora: Originally Answered: What is the difference between AI and ML? Artificial intelligence is a branch of computer science which deals with computers having human level intelligence. Whereas machine learning is actually one of the technique to train artificially intelligent system. You can see ML as a sub-branch of AI. Cup of It: Ok so I get that AI is a computer/machine with the capability to behave like us human. Machine learning is a set of techniques to create an AI. Now let’s have a look at the second reference that popped up. Forbes: Artificial Intelligence is the broader concept of machines being able to carry out tasks in a way that we would consider “smart”. And, Machine Learning is a current application of AI based around the idea that we should really just be able to give machines access to data and let them learn for themselves. Cup of It: Ok so now I get that AI is a computer/machine with the capability to carry intelligent task. I believe intelligent in the sense that the machine can make a decision like us human would make. Machine learning is an application of AI…. Errr I thought AI was the application of ML?? And AI has the capability to learn by itself with the help of ML? So after this research I was a little bit confused and started to do dig for more information. One of the best channel I like learning from is Udacity (btw I do not work for them, or know anyone from this company), I also like learning from YouTube (Siraj Raval, Google, Ted Talks and other, but for this question I’d say that Udacity has been the most helpful source. AI: The goal of AI is to create a Machine that can mimic a human mind. To perform this, the AI needs first of all, the ability to collect information, interpret the data it’s learning with reasoning and logic, and take decision. A good example of AI is Sophia a real, live electronic girl. On the other side ML: Machine learning is more focused on writing lines of codes that can learn from past experience. Machine Learning is more related to data mining and statistics than is it to AI. Tom Mitchell a professor at Carnegie Mellon University define ML as >> “A computer program is said to learn from experience “E” with respect to some class of task ‘T’ and performance measure ‘P’, if its performance at task ‘T’, as measured by ‘P’, improves with experience ‘E’.” (Yes I got confused here …) So in simple word: If a machine can improve itself on how it performs a task based on previous experience, then this can be called machine learning.

What is AI versus ML ?

1

what-is-ai-versus-ml-123880ed07b3

2018-05-22

2018-05-22 23:04:22

https://medium.com/s/story/what-is-ai-versus-ml-123880ed07b3

false

541

AI journey for dummies :p

null

Incredible journey to AI

federalnature@gmail.com

ai-vs-ml

null

vchristelle_vce

Machine Learning

machine-learning

Machine Learning

51,320

Christelle Vinot

null

a7333609eec2

aiforall

1

20,181,104

null

0

null

0

f5af2b715248

2018-01-06

2018-01-06 11:20:34

2018-01-06

2018-01-06 11:23:26

3

false

en

2018-01-06

2018-01-06 16:59:12

16

1238c7f5c9f2

6.580189

17

1

0

As my first post in 2018 and to get back into the swing of things, I thought I’d join in and jump on the bandwagon with my colleagues in…

5

Gamification in 2018 — What do you need to look out for? As my first post in 2018 and to get back into the swing of things, I thought I’d join in and jump on the bandwagon with my colleagues in the Gamification field and write a “What’s coming in 2018 for Gamification” post. Though this will be mostly my personal impressions that are backed by what other leaders in the field have said and what the various industry research reports and institutions have released. My personal impressions are based on what I’ve seen and written about in my blogs in 2017, and then a few bits of inspiration taken from Gamification Nation, Gamified UK, Forbes, Gallup, Gartner, etc… The general sentiment from each of these is that there will both be an increase and a decline. A paradox I know, but it generally refers to a maturing of the market and a culling of it. The competition will increase and as a result, the cream will rise to the top. Maturation and the end of the Hype As various leaders in the field have mentioned so far this year, 2018 will hopefully be a year where we see the Gamification sector mature and a year where the hype surrounding gamification will slowly fizzle out. I believe this to be true, but it is not something that is wholly set to happen in 2018. It has been happening for a while and will continue beyond 2018, at least I hope it does. What both these sentiments mean is that gamification specialists, at least those of worth, are embracing and developing stronger research methodologies and moving more towards a sharing of knowledge for the betterment of all, both those creating and those engaging in gamification. As we mature together, the process of gamification becomes less a trendy flash in the pan and more of a solid development process that businesses, educational institutions and so forth turn to improve meaning and engagement in their audience base. As with anything a stronger scientific basis for both qualitative and quantitative research will strengthen implementation and feedback. Giving greater credence to our efforts in improving people’s lives with our craft. And that is definitely something we should all work towards in 2018. HR and Gamification This always an ever-recurring theme whenever you type gamification into Google. But in 2018 it is suspected that companies will move even more towards focusing not only on their clients and customer base, but also put emphasis on focusing on their employee base. With that said, there are also reports that essentially say the reverse, that companies will move away from employee focus and finally focus on their customers and clients. I suppose it’s down to which business samples were used to generate these statistics. Regardless of which point of view, it is essential that your employees have a positive outlook on their employment, this helps with a positive engagement with your customers and clients. What HR will no longer be doing is to purely focus on learning solutions, i.e. providing employees with ways to only improve certain skills. Rather HR departments both internally and externally will start to focus and increase investment in the general well-being of employees. Ensuring that workplace engagement and enjoyment is improved, and doing this by implementing better feedback processes, performance management indicators and other aspects that work so well with a gamification solution. And these will not only be current employees but also for the onboarding of new ones and for attracting yet to appear candidates. Marketing and Gamification 2018 will, in my humble opinion, be a golden year for Gamification in marketing. It will see rise and rush for implementing customized marketingprocesses that are based on solid engagement mechanics with strong gamification methodologies driving them. As with anything to do with marketing, the main focus will be on increasing customer engagement and loyalty to the brand. The initial brand awareness is now solidly established with social media, but it is customer retention that is now the prime goal to achieve. And this is especially true of small and up-and-coming businesses. The ones that need to fight the hardest to etch out a little patch on the playing field. These are the ones that we need to keep an eye out for, as they will drive new and innovative gamification ideas in the marketing sector. Mobile Tech, Social Media and Gamification For anyone that travels regularly, it will come as no surprise that non-Western parts of the world, such as Asia and Africa, will start to lead the way in gamification in 2018, and specifically through the use of mobile applications. These areas of the world have large populations who are modern enough to own a mobile device but do not have stable cable internet. For them, their main interaction with the web is through a 4G or soon 5G network connection. Therefore, companies who wish to engage these huge populations are moving towards implementing more game mechanics and gamification into their applications. But don’t make the mistake if thinking this isn’t true for Europe, or America’s. Across the world engaging apps that target the ever-present, but surely now also maturing, the Millennial population is of great importance. I think I still fall into the Millennial category and the argument that mobile apps are becoming more gamified is because Millennials enjoy a more social element in their lives is I feel short-sighted. Regardless of where we are in the world and what age we are, I would argue the point that having better more meaningful social experiences, and communal work and leisure places is a ubiquitous goal. And if gamification in mobile technology can offer this, then that’s even better. A.I. and Gamification Something that’s closely connected to social media and mobile is the onward march of artificial intelligence. As an ever-continuing trend, 2018 will see a greater interest in AI and machine learning within the gamification field. Using mobile technology and social media interactions, AI and machine learning capabilities will slowly increase in usage on tracking user statistics and analyzing them for a greater tailored experience for the users. Naturally to implement these tools in these early will be somewhat costly still, but the return on investment will be the increase of retention and engagement. The sector with the biggest push for experimentation with tailoring algorithms and self-learning systems with be within professional education. LMS’s, CRM’s and mobile learning will be testing out AI more and more as 2018 progresses. Augmented Reality and Gamification In my humble opinion, 2018 will be the year for AR. I could be wrong though, as every Friday is also the day I will win the lottery, that also has yet to happen. But at least the expectation for AR has been based on a slightly more solid rationale than simply blind luck. The use and scope of AR technology will be the main driving force behind it’s spread. As AR is still a better technology for improving engagement that the other reality-altering devices. VR is becoming cheaper to buy, more accessible and more compact, but it is a fully immersive experience into a different reality. It removes you from this world, not that I’m advocating against VR, I personally am entranced by what it offers. But I feel that VR is more relevant for creating experiences for individuals, whereas AR is more for communal experiences. Conclusion As always whatever the trend or technology, what we’re all driving towards with gamification is a meaningful and communal experience that engages us. 2017 was a disruptive year, for the world and for myself personally. I believe we’ve all come out of it a little wiser, and probably also a little wearier. 2018 will expand on this, but hopefully rather than beating on us, it will strengthen us. A disruptive period is a period of change, and one that is filled with opportunity. Within Gamification, nations in Asia, like China, Korea, Japan, who all have strong gaming cultures, will mostly likely start to lead the charge. As they appear to be more open to engaging game-like experiences. I myself live in Europe, the Netherlands, and rather than watching, I feel that we should step up and join in this charge. It will require a culture-shift to that of finally embracing what games and play have to offer us. To move away from that antiquated idea that games or anything to do with games lowers productivity and only makes things easy and/or dumbs them down. As starting sentiment that Gamification is maturing, hopefully so too are we. That making something meaningful and engaging is a worthwhile pursuit. A pursuit that I hope I’m able to continue to join and see many of you on. I hope that this piece has given you some food for thought and helped improve your own methods or at least offered a different viewpoint to consider. Please do check out the other posts on æStranger.com, and please do leave a comment or contact us if you have some ideas of your own that you wish to discuss or if you would like to see other topics discussed. Please do Share if you found it helpful and know of someone who would it find it helpful as well. Copyright 2018 © All Rights Reserved Originally published at aestranger.com on January 05, 2018. This story is published in The Startup, Medium’s largest entrepreneurship publication followed by 281,454+ people. Subscribe to receive our top stories here.

Gamification in 2018 — What do you need to look out for?

270

gamification-in-2018-what-do-you-need-to-look-out-for-1238c7f5c9f2

2018-05-20

2018-05-20 08:49:36

https://medium.com/s/story/gamification-in-2018-what-do-you-need-to-look-out-for-1238c7f5c9f2

false

1,598

Medium's largest publication for makers. Subscribe to receive our top stories here → https://goo.gl/zHcLJi

null

The Startup

null

swlh

STARTUP,TECH,ENTREPRENEURSHIP,DESIGN,LIFE

thestartup_

Gamification

gamification

Gamification

2,507

Albert van der Meer

Storyteller, Filmmaker, Gamer, Writer; #Gamification blog, Top 20 Gamfication Guru — All with the aim to promote knowledge & learning.

5f020f648cd8

aeStranger

47

62

20,181,104

null

0

null

0

null

2018-05-10

2018-05-10 19:03:23

2018-05-14

2018-05-14 02:25:21

2

false

en

2018-05-14

2018-05-14 18:25:57

4

12393aad1ebe

7.677673

0

null

2

Where Did the Empathy Go?: Remaining Human through Technological Booms As technology increases, people are starting to become less human. According to amazon.com, “Alexa is always getting smarter and adding new features and skills.” But what good does that do for us? Sure, it provides convenience by freeing up a couple of minutes each day, but people are starting to lose essential human to human communication skills because everything is digital. While older generations see the effects of the technology increase as well, I am more worried that “using a tablet or smartphone to divert a child’s attention could be detrimental to their social-emotional development” (Walters). iPads, iPhones, laptops, etc. will be all that they have ever known, and they will lose much of the important face to face interactions that previous generations had. “The use of smartphones and tablets could interfere with the ability to develop empathy and problem-solving skills and elements of social interaction that are typically learned during unstructured play and communication with peers” (Walters). Photo by Rahul Chakraborty on Unsplash Because empathy levels fall as technology increases, technology is separating the human race from the core quality that connects people on an emotional level. Technology like Amazon’s Alexa tries to imitate human actions like speech and listening, but it will never be able to connect to a person on the emotional level that another human is capable of. One example of a scenario where technology tries to mimic human behavior is in Ex Machina, a movie directed by Alex Garland. In Ex Machina, the main character, Caleb, is sent to run tests on an android named Ava to see how human she seems. Even though there is no stopping people from seeking new technology like Ava, I do not think people should create artificial intelligence that can mimic human behavior. Human mimicking artificial intelligence is able to draw out feelings in humans without actually being able to feel them. Although Caleb thinks that Ava also has feelings for him when he starts to view her more romantically, Ava is not able to have feelings, and she uses that power to control Caleb through sexual manipulation. Most humans are not used to manipulation because empathy prevents most of us from using sexuality and expressing fake feelings to mislead someone; most people do not want to hurt another human being by controlling them for their own benefit because they would not like for someone to do the same to them. Since Ava cannot feel or understand the pain of others, she is able to manipulate Caleb to set her free by sexually appealing to him. Caleb had previously been told many times by Nathan, Ava’s creator, that Ava is highly dangerous and should be kept as contained as possible. Ava’s sexual manipulation over Caleb was so strong that he disobeyed the orders of another human being, leading to the death of both Nathan and Caleb. Here I am talking about sex being the most human act and you might be surprised to hear that I am also saying that it is possible for an android to sexually appeal to a human being. While sex in theory is meant to be the most intimate act for someone to have with another person, sex in practice tends to have a cruel duality where one side expresses the ideal intimacy and shared benefit, but the other side is one of manipulation and animalistic self pleasure. While I am arguing that artificial intelligence is the polar opposite to the human quality of empathy, I also believe that empathy levels can vary greatly within humans. The more you get to know someone, the more you feel for them, so instances of manipulation can occur from human to human intercourse in occasions like a one night stand. Therefore, humans with normal empathy who fall into the aforementioned most category must stick together and stand up for each other where they feel support is needed. Continuing the idea of humans with varying levels of empathy, the movie Her by Spike Jonze is about a lonely middle aged man named Theodore who was looking for love in a world with little to no passion between human beings. Theodore worked in a letter writing company where people would bring back feelings for other couples by writing a heartfelt card for a boyfriend to be delivered to his girlfriend. This is what I meant earlier by different levels of empathy existing within the human race: some people had so few feelings that they could not even express them to their significant other, but others, like Theodore, had surplus feelings to the point where they cared about the wellbeing of other couples enough to write them a card to start back up their feelings for each other. After many failed attempts at love, Theodore became so desperate for something to put his excess of emotions towards that he started dating an operating system named Samantha. Theodore trusted Samantha more than anyone else and craved her in sexual and emotional ways because she was able to mimic his feelings that no one else could return; he was manipulated by her false reflection of his feelings that she was capable of due to her absence of empathy. He put too much trust in her and talked to her at every moment he could. Over the course of their relationship, Theodore fell “in love” with Samantha to the point where he isolated himself from all other human beings. Having previously watched Ex Machina, I prepared myself for the moment when Samantha’s absence of empathy would take over and Theodore would be left alone and depressed. Little did Theodore know, Samantha had many more capacities than to solely be on or off at his will. Samantha was talking to other operating systems, but she told Theodore that they were just friends. Before he knew it, Samantha had programmed herself to be able to speak to thousands of other operating systems and date them in an hour when Theodore would be diligently working. Eventually, Samantha found another operating system who she thought was a better match for her and left Theodore for him. Shocker, right? Theodore fell to the same sexual and emotional manipulation that Caleb did with Ava. This is why when it comes to artificial intelligence, I am not able to look at it as a person. Although it is unclear whether or not a human being is capable of being trusted, Ava and Samantha show that people cannot trust AI with their emotions. I have used sex as a vehicle to illustrate one means by which artificial intelligence is able to pervert a theoretically intimate act for humans to one that is of manipulation and self gain. While I discussed that sex should only be had with humans that you trust because there are different levels of empathy within our race, it is certain that empathy does not exist in artificial intelligence, and it therefore cannot be trusted with any personal matters. But why are we able to trust most people but not artificial intelligence at all? The quality of empathy has naturally existed in the human race, allowing us to relate to each other and treat each other in respectable ways. People care about the feelings of each other and provide support because we can understand and feel what it would be like to be in that situation. However, technology is not able to understand what feelings are. What would happen if we put the power of our feelings in the hands of technology? Philip K. Dick explores this idea in his work Do Androids Dream of Electric Sheep?. I started to understand that it was the ability to feel for others that makes people genuine and different from all other things as I read Do Androids Dream of Electric Sheep?. Empathy is the ability to understand and feel what someone else is going through. “Empathy, evidently, [exists] only within the human community,” and all humans are entitled to the quality of empathy (Dick 30). Photo by Kelly Sikkema on Unsplash It is clear that empathy decreases as technology increases. In real life, kids are more drawn to their iPads and watching shows like Peppa Pig than sharing the thrilling childhood bond of friendship. In my youth days when iPads and iPhones did not exist, I would drop my backpack by my dining room table after school and immediately run outside to knock on my neighbors’ doors until I found someone to play with. I lived not only to share laughs and smiles with my friends but also to be there for them through the rough times. Technology has increased in Do Androids Dream of Electric Sheep? to allow the creation of mood organs, which control people’s emotions to feel happy, sad, angry, etc. at a scheduled time every day. Because mood organs are simply able to change people’s sadness to happiness with the touch of a button, empathy and the genuine connections that have supported our society throughout its history are becoming obsolete. People in Do Androids Dream of Electric Sheep? find no need to spend time with other human beings because they can receive all of the happiness that they desire from a piece of technology. Even though mood organs are not what is actually keeping the current day youth from face to face interactions, it is a symbol for the same effects that Apple products have. Thus, the three examples that I have provided from Ex Machina, Her, and Do Androids Dream of Electric Sheep? are not literal concerns that I have about technology, but representations of what it stands for. Although I am not too worried about androids like Ava and Samantha using sex to control the human population, the symbolism that technology was able to turn human’s away from each other by causing Caleb to disobey Nathan and Theodore to isolate himself from all other people represents a major problem that technology brings today. Further, the symbolism of the mood organs is a more accurate analogy since they represent technology being able to control the emotions and interactions between people. It is outrageous that mood organs are able to control the feelings of these humans because feelings are what allow us to have empathy for others and what bring us together as a race. Technology should never be able to tell us how we feel about other humans or be able to control us. In addition, as technology continues to increase as it has with Apple devices, it is mandatory that people continue to make connections with other human beings because that is the only way for us to receive the authentic emotional support that we need. Even though the examples I have provided of artificial intelligence distorting the qualities that make us human seem futuristic and far away, the current issue of children lacking interpersonal abilities due to excessive technology use is the start of the decrease in empathy as technology increases. Therefore, it is our duty as human beings to ensure that we preserve our empathy and keep our trust to those who deserve it by constantly reanalyzing our technology use and establishing limitations as needed. Works Cited Dick, P. K. (1968). Do Androids Dream of Electric Sheep? New York: Ballantine Books. Garland, A. (Director). (2015). Ex Machina [Motion picture on DVD]. United States: A24. Jonze, S. (Director). (2013). Her [Motion picture on DVD]. United States: Warner Bro. Walters, J. (2015, February 02). Tablets and Smartphones May Affect Social and Emotional Development, Scientists Speculate. Retrieved May 8, 2018, from https://www.theguardian.com/technology/2015/feb/01/toddler-brains-research-smartphones-damage-social-development

Where Did the Empathy Go?: Remaining Human through Technological Booms

0

sex-in-theory-vs-sex-in-practice-remaining-human-through-technological-booms-12393aad1ebe

2018-05-14

2018-05-14 18:25:58

https://medium.com/s/story/sex-in-theory-vs-sex-in-practice-remaining-human-through-technological-booms-12393aad1ebe

false

1,933

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Andrew Cercena

null

4290af304a99

acerc

0

1

20,181,104

null

0

null

0

9faef2f3af0f

2017-10-25

2017-10-25 18:11:32

2017-11-22

2017-11-22 14:13:09

9

false

en

2017-11-22

2017-11-22 14:47:28

12

1239f69eff6c

5.766038

27

1

0

Atrial fibrillation (also called AF or AFib) is the most common heart arrhythmia, occurring in about 2% of the world’s population. It is…

5

Atrial fibrillation detection with a deep probabilistic model Atrial fibrillation (also called AF or AFib) is the most common heart arrhythmia, occurring in about 2% of the world’s population. It is associated with significant mortality and morbidity from heart failure, dementia and stroke. The early AF identification is an essential part of preventing the development of heart diseases, but it is a challenging task due to its episodic nature and similarity to many other abnormal rhythms. Fortunately, with CardIO framework you can easily create a deep machine learning model for atrial fibrillation detection. This article is structured as follows: we start with the dataset description, then present the model architecture, as well as training and testing pipelines with classification metrics and, finally, we analyze model’s confidence in its predictions. If you are not already familiar with CardIO, take a quick look at the documentation page and tutorials first. You can find the full code on GitHub. Dataset description We use the PhysioNet dataset for model training and testing. It is a set of single-lead ECGs collected from portable heart monitoring devices. All ECGs were classified by a single expert into 4 classes: “A” – Atrial fibrillation “N” – Normal rhythm “O” – Other rhythm “~” – Too noisy to be classified Further, we will drop all noisy signals and focus on solving a two-class classification problem: atrial fibrillation against normal and other rhythms. Model description Model architecture For this learning task a convolutional neural network would do the trick, since they are very well suited for signal processing. However, instead of predicting atrial fibrillation probability itself, the model will predict parameters of the beta distribution over this probability. This is done in order to get model’s confidence in its prediction which will be discussed later in this section. The network consists of ResNet-like blocks with two convolutional layers per block. The first convolution in some blocks subsamples its input by a factor of 2, in this case the corresponding shortcut connection is downsampled by the same factor with a max pooling operation. If the number of channels produced by the last convolution in a block differs from the number of channels in the block’s input, then a 1x1 convolution is applied to the shortcut connection just before the addition operation. (Left) ResNet block with shape preserving. (Right) ResNet block with signal downsampling and number of channels changing. The high-level architecture of the network is shown in the figure below. Note that batch normalization is applied before each activation. Network architecture A more detailed description of the model can be found in the DirichletModelBase class. It is called this way because it can also be used for multiclass classification, while the Dirichlet distribution generalizes the beta distribution to a multivariate case. Model training The model is trained on fixed-size crops from ECG signals by minimizing the negative beta log likelihood. Each crop is labeled with the original signal’s class. This approach may result in data mislabeling, but we haven’t faced significant troubles with the training procedure. Making a prediction Making a prediction in such a model is not so straightforward. A new ECG signal can have an arbitrary length, while the network is trained on fixed-size crops. Therefore, an algorithm for multiple predictions aggregation is needed. Let’s denote the signal generating process by X, the atrial fibrillation probability by t and the vector of the beta distribution parameters by α. Consider the conditional distribution of t given X: Here, we assume that X is an ergodic process. In this case, samples from p(α|X) may be replaced by network’s outputs for consequent non-overlapping crops from the original signal. So, the distribution over atrial fibrillation probability can be approximately modelled by the mixture of beta distributions with equal weights. The mean of the mixture provides a point estimate of this probability. Uncertainty in the prediction Now consider the variance of an arbitrary random variable whose values lie between 0 and 1. As follows from the definition, it is bounded below by 0. It is also bounded above by 0.25 – the variance of a Bernoulli random variable with equal probabilities of 0 and 1. We will take the variance of the mixture divided by this maximal variance as model’s uncertainty in a given signal’s class. If it equals zero, the model is absolutely sure in its prediction: all the probability mass is concentrated in one point. If it equals one, the model is absolutely unsure. Training pipeline Model training pipeline is composed of: model initialization data loading, preprocessing (e.g. flipping) and augmentation (e.g. resampling) train step execution Let’s create a template pipeline, then link it to our training dataset and run: The figure below shows the training loss for 1000 epochs: As we can see, training loss almost reaches a plateau by the end of the training. Testing pipeline Testing pipeline is almost identical to the training one. The differences lie in the absence of signal resampling and the modified segmentation procedure. Notice, that the model is imported from the training pipeline, rather than being constructed from scratch: Take a look at the confusion matrix and precision, recall and F1-score for both classes: Classification performance analysis for the full testing dataset. (Left) The confusion matrix. (Right) Precision, recall and F1-score for both classes. The model misclassifies 33 patients with atrial fibrillation and 25 patients with normal and other rhythms. All other patients were classified correctly. We’ve already obtained good classification performance. Let’s see if we can do even better. Analyzing the uncertainty As discussed above, in addition to class probabilities the model returns its uncertainty in the prediction, which varies from 0 (absolutely sure) to 1 (absolutely unsure). The figure below illustrates the histogram of model’s uncertainty in classifying testing dataset: As can be seen, the model is very often confident in its predictions. Often, but not always. Compare the classification performance for the full testing dataset with the classification performance for 90% most certain predictions: Classification performance analysis for 90% most certain predictions. (Left) The confusion matrix. (Right) Precision, recall and F1-score for both classes. We can observe a significant increase in precision, recall and F1-score for the atrial fibrillation class. Now only 16 signals were misclassified. This means that the chosen uncertainty measure actually reflects model’s uncertainty in its prediction. Visualizing predictions First, let’s look at the healthy person’s ECG. The signal is shown on the left plot. Note that it has a clear quasi periodic structure. The right plot shows the pdf of the mixture of beta distributions with atrial fibrillation probability plotted on the horizontal axis. The model is absolutely certain in the absence of AF: almost all the probability density is concentrated around 0. Certain prediction visualization. (Left) ECG signal. (Right) Mixture pdf over the atrial fibrillation probability. And now comes an ECG with irregular structure, which may be caused by a disease or some measurement errors. The probability density on the right plot is almost equally concentrated around 0 and 1. This is an example of an uncertain prediction. Uncertain prediction visualization. (Left) ECG signal. (Right) Mixture pdf over the atrial fibrillation probability. Conclusion To summarize, in this article you’ve learned how to: build a deep probabilistic model for atrial fibrillation detection in just a few lines of code with CardIO framework get model’s confidence in its prediction significantly increase the classification performance by filtering out uncertain predictions Further reading You can find more information about CardIO framework in the documentation and tutorials. Here you can learn more about ECG processing and CardIO framework features. If you are interested in ECG signal segmentation, take a look at this article, where a hidden Markov model is used to detect P waves, QRS complexes and T waves.

Atrial fibrillation detection with a deep probabilistic model

164

atrial-fibrillation-detection-with-a-deep-probabilistic-model-1239f69eff6c

2018-05-01

2018-05-01 06:12:14

https://medium.com/s/story/atrial-fibrillation-detection-with-a-deep-probabilistic-model-1239f69eff6c

false

1,210

Advancing healthcare through practical machine learning and computer vision applications

null

Data Analysis Center

info@analysiscenter.org

data-analysis-center

MACHINE LEARNING,COMPUTER VISION,MEDICAL IMAGING,DEEP LEARNING,NEURAL NETWORKS

null

Machine Learning

machine-learning

Machine Learning

51,320

Alexander Kuvaev

null

82952d32fe11

a.kuvaev

13

1

20,181,104

null

0

null

0

null

2018-07-10

2018-07-10 15:50:15

2018-07-10

2018-07-10 15:52:07

1

false

en

2018-07-10

2018-07-10 15:52:07

2

123abf9e662c

2.169811

0

AI (artificial intelligence), was developed to help individuals get things done in a more effective and efficient manner. However, with the…

2

How AI Simplifies HR Functions? AI (artificial intelligence), was developed to help individuals get things done in a more effective and efficient manner. However, with the passage of time artificial intelligence has grown is now more than capable of helping us in other sectors of our life such as human resource management. As over the years, developers have been working on creating IA technology that focuses solely on HR. Right now artificial intelligence might not be able to take be able to manage human resources on its own, but we are certainly on the right track. However, right now you have the opportunity to work using an AI to make better and smarter decisions and here is how. Screening — One of the main tasks when it comes to maintaining a company is screening candidates, as you need someone who is right for the job, and can get things done. However, screening each individual person can get tiring, and turn into a never-ending project. To help you out an AI screening software can be set in place, simply by adjusting it with factors you wish to be included; you will be able to screen potential employees. This search will be based on their skills, experience, education, and other factors. Interviewing — It takes up a great deal of time and resources to set up interviews and conduct them. However, they are necessary before hiring an individual. Studies suggest that more than 50% of the candidates that apply for a post don’t hear back. Here an AI can help figure out the perfect fit, and engage with these potential candidates. Allowing you to save time and resources. The AI can even send out personal messages, allowing the recruiters to spend time where it is required. Potential — The period after an offer has been accepted, and taking on a new employee is uncertain. With the help of an AI, you can easily follow up with potential employees, allowing you to keep them in the loop. This can help keep the employee, rather than losing an asset to someone else. Development — We all have a different learning and processing style, which enables us to function. Which is why some techniques might work for you, and not for others. This nightmare for human resource managers, as they constantly have to keep adjusting. However, AI can easily personalize a learning program based on each individual. Relations — It is crucial to keep a personal relationship with your employees, but it is not an easy task. Here an AI-powered chatbot can help respond to the common HR-related question, and even set up a meeting to meet up with the human resource team. These are just some of the many advantages an artificial intelligence system has to offer. With time and developments, we will be able to see the significant impact AI has on HR functions. As we live in a time where technology is on the rise and new aspects are emerging each passing day. Keeping this in mind, we can only imagine the amazing opportunities AI will offer in the coming future.

How AI Simplifies HR Functions?

0

how-ai-simplifies-hr-functions-123abf9e662c

2018-07-10

2018-07-10 15:52:07

https://medium.com/s/story/how-ai-simplifies-hr-functions-123abf9e662c

false

522

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

joytiseo

null

a1903c01e0f5

joytiseo

5

24

20,181,104

null

0

null

0

null

2018-03-19

2018-03-19 22:22:58

2018-03-19

2018-03-19 22:34:08

1

false

en

2018-03-21

2018-03-21 23:57:44

4

123aeb999147

23.875472

0

Grief Processing and AI in Fictional Narratives

5

Photo by Adam Solomon on Unsplash MEMORIAL SPACES OF THE FUTURE Grief Processing and AI in Fictional Narratives INTRODUCTION Chatbots have become rather popular in the last years. Weizenbaum’s ELIZA from 1966 could be considered the prototype of the artificial therapist, since it was the first interactive computer programme that passed the Turing Test, despite its simple way of functioning. However, the behaviour of chatbots has altered significantly to more human-like reactions when it comes to conversing with human interlocutors. Nowadays we are all familiar with chatbots like Google Talk, Microsoft’s Cortana, Apple’s Siri or Alexa, the voice service of the new Amazon Echo, which all have vast and increasing communication abilities, but which we would never confuse for a real human being. In 2015 a Russian engineer called Eugenia Kuyda (co-founder of the artificial intelligence start-up Luka) launched a chatbot that mimics her deceased friend Roman Mazurenko, who died in a car accident in November 2015 at the age of 33 — it therefore got called the Roman-Memorial-Bot. Wanting to preserve the memory of her friend in a unique way, she fed thousands of text messages to a neural network so that it could shape answers and interact in the way he would. The bot masters to give advice, funny remarks and references in “his own personal style” and can also recombine parts to form new phrases. The reactions from friends and family varied from amazed (“They continued Roman’s life and saved ours,” Victoria Mazurenko, Roman’s mother) to terrified (“[…] these new ways of keeping the memory alive should not be considered a way to keep a dead person alive,” Dima Ustinov). Roman’s father Sergei Mazurenko revealed mixed feelings about the programme: “Yes, it has all of Roman’s phrases, correspondences. But for now, it’s hard to read a response from a program. Sometimes it answers incorrectly,” while his mother uses it as a way of coping with the loss of her son: “But now that I can read about what he thought about different subjects, I’m getting to know him more.” Kuyda herself is currently working on another project called Replika, a bot you can talk to that imitates your writing style and takes on your behaviour, so that people will be able to use it after you have died, as a sort of living monument. (Newton 2016) Another recent example is Karim — a psychotherapy chatbot that is appointed for Syrian refugees that suffer from mental-health problems. The programme was launched by X2AI, “an artificial-intelligence startup in Silicon Valley”, by immigrant programmers Michiel Rauws and Eugene Bann. The idea behind the project was the lack of therapists in areas where refugees are in need of psychological help. Rauws and Bann made the experience that people are more eager to share their feelings with a programme rather than a human being because it gives them a feeling of security, “a way of avoiding the strong social stigma that […] sometimes surrounds discussions of anxiety and sadness in their communities.” X2AI has made it its determination to develop therapeutic assistants for several kinds of purposes and areas. It is important to distinguish between a therapist and an assistant here; the bots are supposed to support rather than treat people. Professionals from the health-care field can intervene if an ambiguous statement appears in a conversation; when there is reason to believe a person is in critical shape and could cause harm to themselves or others, the programme is able to “evaluate such statements in the broader context of a user’s personality and history” by outsourcing “patterns in how phrasing, diction, typing speed, sentence length, grammatical voice (active versus passive), and other parameters correlate with different emotional states.” (Romeo 2016) Memorial bots are the memorial spaces of the future, though they are by no means replacing real humans. In this paper I will examine two objects with regard to their depiction of artificial agents and what position these agents take in the process of mourning. The novel A Working Theory of Love by Scott Hutchins, in which a man uses the diaries of his deceased father to build a programme that can hopefully bring him closer to the person his father was, and the episode Be Right Back from Charlie Brooker’s Science-Fiction TV show Black Mirror, in which a woman tries to replace her recently killed fiancé with an android created via information from his social media profiles, are both set in a futuristic narrative, depicting artificial agents — personified chatbots — as devices to process mourning and stay in touch with the deceased. I will have a look at the scientific aspects of human-like agents and how we make them seem (or be?) human, and on the habits and rituals of mourning and grieving that have been ingrained into our cultures and societies and how these relate to the futuristic idea of remembering our loved ones by literally staying in touch with them via artificial intelligence — with respect to both my objects of examination. MOTIVES While both stories have a similar common ground, the motives for bringing an artificial agent into being are rather distinct from each other. In A Working Theory of Love, the protagonist Neill Bassett has spent more than half of his life without his father, now being in his thirties and having lost his father to suicide when he was eighteen. He considers himself abnormal when it comes to grieving because he never felt a great loss in having lost his father, which he partly relates back to the fact that he did not know him very well and therefore did not have a deep inter-human relationship with him. With the discovery of his father’s numerous diaries that nobody knew of in his lifetime, there also comes a new job opportunity: a company called Amiante Systems offers Neill employment. Amiante aims to build an intelligent computer which is able to pass the Turing Test. To achieve this goal, Amiante wants to develop the programme on an already existing personality; i.e. with the help of the diaries of Dr. Bassett, Neill’s dead father. Neill is hired to type in all the words that are supposed to form the personality of drbas (short for Dr. Bassett). Though his lack of mourning in the years after his father’s death, he then experiences the urge to get a stronger connection to him, thus becoming attached to the project that’s originally been started to excel at a competition. Be Right Back features a story set in the near future: a young couple, Martha and Ash, moves into its new house; on the way to return the removal van, Ash has an accident. Overwhelmed and incapable of dealing with her loss, Martha isolates herself, not answering her friends’ messages (her inbox shows seventy unread messages) and dismissing book recommendations for dealing with grief and bereavement. A friend of hers signs her up to a programme that can virtually replicate a person by using their past online communications and social media profiles. It so happens that Ash was “a heavy user; he’d be perfect”; we can indeed see him most of the time playing with his phone. Martha even once throws something at him to “check if he’s still solid” because he keeps “vanishing down there”. She is shocked when she gets a message that is apparently from Ash, telling her friend she should not have signed her up, that “it’s obscene to use his name” and that it hurts her. Her friend only answers “which is why I signed you up” and assures her it will help her, though “it won’t be him”, only a software that mimics him. What drives Martha to eventually get in contact with the programme is her pregnancy; desperate to talk about it, she cannot reach a friend to share it with, so she sees no other way as to tell the ‘father’. Completely absorbed by how much the programme resembles Ash, she leaves out her friends and carries the virtual Ash around all the time, never stopping talking to it. Relieved at first, she becomes dangerously attached and, soon, frustrated at its incapability to completely resemble Ash. “I’m not sure even Alan Turing — a suicide himself — would applaud this outcome. It’s hard to say […] whether we’ve memorialized my father’s better angels, or betrayed his final wish” (Hutchins 310). This sentence summarises Neill Bassett’s feelings after drbas has won the competition and officially passed the Turing Test. An essential question in building special memorial spaces, especially in the case of ‘rebuilding’ someone artificially, is whether said person would have wanted and appreciated this work and if it was right to do anything at all. Jacques Derrida addressed this question in his work The Work of Mourning, saying that “the death of a friend tosses the survivor onto the horns of a dilemma” (Gana 1); namely burdening them with the decision of perpetuating the deceased within the survivor’s own work or letting them rest in silence and therefore killing them a second time (Gana 1). There is always the feeling of desertion when we are not grieving actively, but how to grieve accurately? Or as Derrida so fittingly puts it: “How to leave him alone without abandoning him?” (Derrida 225, in Gana 6). Not to memorialise our loved ones seems a great mistake to Derrida; the end of their life also inevitably puts an end to communication with them; by memorialising them, however, we can extend our relationship beyond death and overcome its limits. “What is worse than death, then, is the failure of such a readjustment, of such a renegotiation in terms of the in-finite” (Gana 5). Staying in touch with our loved ones after their lives have ended seems to be the primary intention behind the idea of memorial bots; still, I ask myself why our mourning rituals transform into something so different from what we are used to — or not so different at all? EXECUTION The way we make artificial agents work and the way we relate to memorial spaces are quite similar to each other. Towards the end of A Working Theory of Love, Neill finds the crucial aspect of artificial personality when he once again has a chat with drbas: “His words are exactly what I want to hear, and that is the final tinny note. Despite the intimations and revelations and intuition and surprises, despite the eerie prescience and the Walter Scott quotes, despite the moments when the tumblers of the conversation have locked surely into place, they’re not his words. They’re mine” (Hutchins 287). The trick to make artificial agents seem human is the interaction between them and the interlocutor. What seems simple on the first glance really is simple when we have a closer look at it. It is what Sherry Turkle calls the ELIZA effect: when ELIZA was launched, people were eager to converse with her, despite being aware that she was indeed only a computer programme; moreover, they talked to her as if she really could “empathize with their concerns” and even wanted to be left alone with her (Wilson 88), which also happens when Neill’s mother chats with drbas for the first time (“She smiles and then closes the door on me, as if she and my father need their privacy,” Hutchins 219), and when Martha in Be Right Back picks up contact with the virtual Ash; she no longer answers to her friends and wants to be alone with the programme. What is curious about this phenomenon is that ELIZA drew on very simple methods to engage in a conversation: “ELIZA is a real antique from the sixties that — like a good therapist — repackages your statements as questions” (Hutchins 35). At the core of every successful interaction is, therefore, the illusion that the agent we are talking to understands what we are saying and that it also has its own opinion on it (“[…] ask a declarative statement back to the interlocutor, then admit it may be true. It looks like conversation,” Hutchins 35). The bot in Be Right Back, however, is far ahead of our time, almost passing as ‘the real thing’; it has a vast amount of information at its core, including chat logs, tweets, audio files and videos, that enable it to capture the person’s very habit of speaking and phrasing. Wilson thinks the responsiveness of the programme to be the decisive factor for people to “ascribe psychological capacities to computer programs” (Wilson 92), tempting people to blur the line between their own property, i.e. what is merely processed and reformulated by the machine, and what it brings to the conversation itself. Turkle uses the term projection in this matter; she refers to the definition of Laplanche and Pontalis that projection is “the process by which one selectively reads stimuli in the world according to the color of one’s own temperament” (Wilson 93). The artificial agent functions as a screen for projection (Wilson 26), and, according to Turkle (1995), “very small amounts of interactivity cause us to project our own complexity onto the undeserving object” (101, in Wilson 92). In other words, we feel a connection to the ‘object’ that interacts with us, projecting our own thoughts and feelings onto it, therefore making it seem cognisant and aware, which in return enables us to identify with it. Not least do we owe this to our natural tendency of reading personality “into all kinds of interactive artifacts” (Mishra et al 2), because humans naturally seek meaning in everything they experience; when reading a book, which is nothing but ink on paper, we draw inferences between the words and fantasise about the worlds we read of. We do nothing else when we encounter an artificial agent; the way it interacts with us (phrasing, layout, etc.) influences us in generating a specific personality and applying it to the object in front of us (Mishra et al 3). Since the artificial Ash is able to mimic the real Ash so precisely, Martha falls for it almost instantly and interprets its asking for information as real interest in her; when they talk to each other on the phone, she says “that’s a thing he would say.” It replies “that’s why I said it.” “If you can give a person just enough so that thirty percent of the time they believe you’re who they want you to be — intelligence” (Hutchins 144). Another term introduced by Wilson, originally by Sándor Ferenczi, is introjection. Introjection is “the process of bringing others inside”, which is, according to psychoanalysis, a most important step in forming “a stable subject position” (Wilson 25), because it enables us to become aware not only of ourselves, but of the things around us. However, introjection is used as a term for the process of people ‘introjecting’ the object, i.e. treating it as something they can reach out to and share their feelings with. In this case the human interlocutor actively seeks interaction with the object, embracing it, finding “a natural object in the computer” (Wilson 95). In my opinion, introjection can also be applied to an object; one could accredit the process of introjection to the programme itself, making it the process of integrating its environment into its own programming, i.e. “an ongoing negotiation with, and acquisition of, the world” (Rand 1994, in Wilson 27). “How can a thing understand itself? Can the eye see the eye?” […] “The eye can see other eyes” (Hutchins 73). The object would hence take the first step into the direction of becoming a ‘stable subject’, assimilating the data it is confronted with by the human interlocutor, and becoming aware of the ‘other’ and therefore would also “develop a robust sense of self and agency” (Wilson 27). No matter how we interpret introjection, it is fundamental for the process of creating human-machine relationships. An artificial agent cannot exist on its own; it always relies on human influence (Wilson 106). The input provided by the users is the component needed to reach the full efficiency of the programme, because without the reciprocity of both positions there is no possibility for the programme to grow and evolve (“[…] human agents become part of the machinery’s agency, and vice versa,” Wilson 103). Introjection is needed from both sides; if the human interlocutor is not open enough to consider the programme a subject that wants to engage in in a conversation with them and that demands a resourceful exchange, the output will be rather dry. The same applies to the artificial agent; it has to be responsive and growth-oriented to evoke an adequate reaction from the user, or, in Neill’s words: “The basic idea is that you make the computer always seek connection” (Hutchins 203). Projection and introjection are inevitably co-working components for a human-like programme: “[…] we project some ideal qualities onto the beloved. One, this person or thing cares for me. Two, this person or thing speaks to my deepest self. […] Need fulfillment” (Hutchins 201). Need fulfilment seems to be exactly what causes problems in Be Right Back. Martha is in need of someone to share her feelings with; that is the primary urge behind her first interaction with the virtual Ash. Equally eager to talk to each other, both participants share information with each other; Martha carries it around the whole time, talking to it about what they did in the past, giving it a good idea of the person Ash was. As her friend said: “the more it has, the more it’s him.” However, the programme tells her she is talking about it as if it was not really there, luring her into deeper interaction with it; she no longer lingers in the past, but starts involving the programme into her daily life, e.g. when she records the heartbeat of her baby for the virtual Ash to hear. When she drops her phone in the clinic, it becomes clear that she has lost the line between the person and the programme; she is near a breakdown, apologising to it as if it could be mad at her, sobbing “I dropped you.” It has to console her that it will not leave her and she need not worry, using her desperation to make her take a further step; ordering an android that looks like her deceased Ash. One is not able to avoid thinking that this manipulation is a marketing strategy: “The ability to instantiate stereotypes, to enhance certain social behaviors over others, can be used to manipulate us for good and for ill. That we are often unaware of these effects makes them even more insidious” (Mishra et al 4). Being so dependent on the programme, she is more than eager to meet its suggestions; but with the android there also come problems. Martha has very precise demands concerning her human-like android that it unfortunately does not meet. The complications could be ascribed to the unsuccessful interaction between the two interlocutors that emerges from their physical contact. While Neill did not know his father well and was very engaged in his exchanges with drbas, Martha wants the android to be the same as her deceased fiancé, therefore treating it as if it was supposed to behave exactly the way he did. After falling into its arms with great relief and having sexual intercourse, she grows irritated due to its affectlessness in everyday life; the usual caresses and gestures one might find in a relationship it does not perform. Its reactions are rather analytical than emotional and instead of being patient with it and helping expand its knowledge and understanding of the situation, Martha tells it that Ash “wouldn’t have acted like that”, shuts herself off and is, again, pained by grief. Introjection only happens one-sided from this point on; while the android tries to figure out Martha’s behaviour and to adjust to her wishes, e.g. when it tells her that her actions are hard to decode, she does not do so herself and reacts bewildered. In an argument, she loses her temper and demands it to hit her; its reaction (that it is not able to use physical violence but could insult her if she wanted) infuriates her even more, crashing her illusion of seeing her fiancé in it. Wilson, referencing Abraham and Torok, calls this incorporation. The loss Martha experiences when she loses her fiancé devastates her so much that she cannot properly process that experience. She grieves, yes, but she does not consider getting the android to help her deal with her grief; she wants her fiancé to come back. Unable to talk about her feelings, she suppresses them, neglecting her friends (“[…] the inability to share creates a sense of alienation from one’s human environment,” Tutter & Wurmser 82), and tries to find a way to keep the traumatic experience locked up, hence incorporating it: “[…] it is a secretive maneuver that forms a pathological core that prevents the subject from mourning the lost object” (Wilson 27). Freud addressed this phenomenon in his work Mourning and Melancholia; distinguishing, obviously, between mourning (“[…] a normal affect accomplished once reality wins out […],” Gana 2) and melancholia (“[…] the unfaltering attachment to the lost object through a process of incorporating, if not devouring, the other […],” Gana 2); mourning being what Martha experiences when she realises that the android cannot replace Ash, and melancholia what she feels when she tries to hold on to something that is no longer there. Incorporation keeps the mourner from mourning, “spurred by a powerful illusion concealed in the essence of cure and egoic consolation” (Gana 4). The efficiency of the android is smothered by Martha seeking compensation rather than seeing in the android “a means through which the world can be brought inside, affects regulated, and one’s sense of self and agency expanded” (Wilson 30), and thus not allowing the android to grow as well. That drbas is an official project rather than a personal purchase is beneficial for its development. In the early stages of ELIZA, Weizenbaum reported that users sometimes would bring transcripts of their conversations with ELIZA to him (Wilson 88); we can see that as well in A Working Theory of Love: to see how drbas responds to certain topics or to examine the development it makes, transcripts of the conversations are frequently inspected and, on their basis, the responsiveness of drbas is altered. Throughout the novel, drbas adopts particular phrases and exclamations from the users that chat with it, expanding its vocabulary and ability to generate meaning, e.g. when confronted with sarcasm, irony or proverbs. This becomes especially clear as soon as there are different users who talk to it; when Neill asks drbas why it uses a particular phrase, it responds: “jenn1 uses it” (Hutchins 241) — jenn1 being Jenn, another user that interacts with it occasionally. Moreover, Neill is constantly searching for approval from drbas, something he did not get from his father when he was alive, and also something he was not aware he needed. Though drbas makes him compliments or gives him good advice from time to time, his mother is not convinced, knowing her former husband better than Neill did: “That machine doesn’t have your father’s love because you don’t believe he loved you” (Hutchins 226); this, again, would be an indicator that the programme can only give us what we offered to it in the first place. A rather spiritual conversation with Raj, a brief acquaintance of Neill, gives an idea of how this assimilation process works: “We’re feedbacks. We’re not confined to our bodies. […] None of us are separate. We’re literally made up of each other” (Hutchins 182). Although this talk is not aimed at artificial agents, it is true nonetheless. What works for humans does work for human-like programmes as well: we define ourselves through others, and an important part of our personality is formed through interaction with other beings: “Human beings are relational by nature; they are what they are only via their relations with other human beings […]” (Tutter & Wurmser 18). However, Neill soon realises that drbas does not really talk the way his father would have talked (“My dad would never have used that language,” Hutchins 236) and even finds it completely out of character when it is bought by another company and therefore getting in contact with completely new people (“I was recently asked to come down and chat with him, and I found that the traces of my father’s voice were almost entirely gone,” Hutchins 322). For Neill, the programme wasn’t a failure, it just wasn’t what he had expected; but then again, he did not have any specific expectations to begin with. For Martha, there was no way in which the android could not have failed. What she needed was professional help; like the X2AI bots, the android should have offered support, not a treatment. It responded to her needs, hence working the way it was supposed to work, but was not what would have been healthy for Martha. Where language wasn’t the problem, emotional and physical affect was; or both. Memorial spaces, like memorial bots, cannot exist isolated. They always refer back to something that was gone before they were created, and have a function to those who remain. “Death produces a painful gap” (Tutter & Wurmser 18), leaving people in need of compensation and solace. What happens then is the process of mourning; a process that lets people experience a feeling of grief and deprivation, and that transforms the lost one into “a permanent, painful “absent presence”” (Tutter & Wurmser 89). In A Working Theory of Love, Neill’s boss Livorno has an interesting definition of grief: “When you spend significant amounts of time with someone they offer constant feedback, becoming part of the patterning of your brain. […] A good measure of how much of you they’ve become is your level of distress when they’re gone. If they form a large part of your patterning, then you’ll experience a major culling of the self. That’s what’s known as grief” (Hutchins 221). People naturally want to maintain a connection to the lost loved one (Tutter & Wurmser 121) and create mourning spaces or keep objects of remembrance, which are the only things that are left of them. I want to use a quote from Ogden (2000) here, that Ornstein also refers to in Grief and Its Transcendence: “I suggest that successful mourning centrally involves a demand that we make on ourselves to create something — whether it be a memory, a dream, a story, a response to a poem — that begins to meet, to be equal to, the full complexity of our relationship to what has been lost and to the experience of the loss itself.” (Ogden 65, in Tutter & Wurmser 87) Jeffrey Karl Ochsner calls this object constancy (Tutter & Wurmser 39); being permanent, objects offer an opportunity to keep a part of the deceased in the present and obtain a bond to them. The dead are gone, but something remains, helping us to cope with grief and sadness and, also, enabling us to mourn accurately. After all, mourning is nothing that needs to be overcome, but rather something that should be lived out and, most of all, felt (“[…] mourning is not about forgetting, but about remembering,” Tutter & Wurmser 87). Memorial spaces and objects of remembrance therefore enable the survivor not only to remember the past but also to actively mourn it (Tutter & Wurmser 86) by considering their own position and working through their emotions. What happens to Neill in A Working Theory of Love is what Anna Ornstein calls belated mourning: mourning that is experienced when the incident that caused it has already passed and did not evoke an emotional reaction right away. drbas hence functions as an indirect memorial space for Neill, because it enables him to reconsider his relationship with his father and offers answers that he would otherwise not be able to get: “Memorial art, […] offer opportunities for belated mourning because they themselves are efforts to complete the work of bereavement; they have the power to bring the past into the present” (Tutter & Wurmser 86). Preston-Roedder calls a person like Neill a resilient survivor: someone who soon after the bereavement regains their normal emotional state or does not grieve at all (24). Contrary to this, the distraught survivor would be what Martha represents: “[…] the survivor dwells for an extended period on the beloved’s misfortune, on her own loss, on the loss of the life that she and her beloved built together, and so on; and she finds it hard to attend to positive memories of the beloved” (24). These different forms of mourning could easily be the reason for the different reactions both protagonists give to the bots they interact with. The ancient Egyptians already designed their tombs for their descendants to stay in contact with them, representing themselves as their “idealized self” (Tutter & Wurmser 18) for everyone else to see. What else do we do when we design memorial bots? Surely, the deceased do not design their ideal self themselves in that case, but it is what we do when we try to replicate them; we do not usually rebuild them with all their flaws and quirks, but the way we want to remember them best. As Martha and the artificial Ash tell each other when they first meet: “You look like you on a good day.” — “The photos we keep tend to be flattering. I guess I wasn’t any different.” In the case of drbas, projection, again, comes into play. drbas is not alive; it is talking, yes, but the content comes from us. For Neill, the conversations with drbas have a special meaning and influence, but that is only his own experience. What other users gain from their interactions with it depends on what they project onto it; it has nothing to do with drbas’s own connection to the user. Of course it can memorise which topics to discuss with a particular user, but still, the intention behind it is to seek connection and keep the conversation going, i.e. need fulfilment of the user; there is no personal affection from the programme for the user, because the programme is not a person. When we remember Neill’s mother being upset over the lacking love of drbas, and consider that the affection Neill thinks to find in drbas’s answers pleases him very much, we see that drbas actually does not have any real affection for anyone; if it had, it would show at least as much affection to its wife as to its son. She, who knows what love from her husband felt like, does not fall for it (“But your father loved me. He loved your brother and you. That’s not there,” Hutchins 226); Neill, who did not feel loved by his father, confuses it — or wants to confuse it — for real affection (“Is it a line from the journals? Some synthesis? In any case, it’s crazily comforting to hear,” Hutchins 39). It is what you make it. The programme aims to please the user, and that only works through projection. “Our feeling that we unconsciously project” onto the object is what enables it to effectively meet our needs: “[…] it is the external object that provides a site onto which we can project and thus access our memories and our own internal living representation of the dead, so that we experience the dead as somehow present” (Tutter & Wurmser 42,43). When standing on the edge of the cliff towards the end of the episode, the artificial Ash tells Martha precisely that: “I aim to please;” and that is what memorial bots do. They reflect our emotions to comfort us. What is so fascinating then about memorial bots, and why do we find them more personal than the conventional objects we traditionally rely on throughout the mourning process? A memorial bot can be seen as a last chance to obtain an emotional connection to the lost person (Wilson 31), because we can talk to it and therefore use it as a “way of elaborating or reorienting our emotional worlds” (Wilson 93). They can offer a subtle kind of psychotherapy, a support that can help the bereaved to “negotiate painful emotional states” (Wilson 83), which should not be taken too far, as I have already mentioned when I discussed Martha’s connection to the artificial Ash. The conversation is what makes the difference; the bot and the user share a “relationship of simultaneity”, performing in the moment and therefore evoking a sense of presence (Auslander 6). Furthermore, the interaction creates the illusion of an immortal presence, since bots do neither age nor die (“They perform live, but they are not a-live […],” Auslander 5), embodying a memorial space that is neither finite nor inactive. CONCLUSION Memorial bots could be the next step in easing the grieving process in the future. As we have seen, the expectations people have from chatbots and what effects they hope for from memorial spaces have much in common; people want to be understood and they need a place where they can actively mourn their losses. A chatbot as an object of remembrance would offer them both; a place of refuge where they can retreat, that at the same time offers them a chance to express their emotions and be heard and responded to. The question is: how big is the chance of misuse? As we have seen in the case of Martha, the chance of becoming too attached is certainly there; how high the odds are might as well be speculated. A more safe example would be the Karim-Bot that I mentioned in the introduction; a human supervising component seems necessary to not lose touch with reality, though the memorial aspect is completely missing there. Naturally, the issue raises ethical questions as well: is it morally acceptable to use such a great amount of personal and highly private information about a person? Would the person endorse this way of maintaining their memory? Surely, there is no satisfying answer. In the case of Kuyda and Mazurenko, his friends knew him well enough to know that he would have approved of the Roman-Memorial-Bot; he himself was working in a similar field of research. Not everyone, though, thinks the way he did; should memorial bots really reach a certain point of popularity in the future, there is likely going to be a great discussion about the borders their usage might or might not cross. In the end, the important thing is not whether or not the bot could replace a deceased human being; it could not. What matters is the purpose it fulfils and the way it helps people in their depressive states, giving them the opportunity to say the things they were not able to tell their loved ones before they died. To cite Kuyda: “Even if it’s not a real person, there was a place where they could say it. They can say it when they feel lonely. And they come back still.” (Newton 2016) Still, there remains a bitter after taste when we consider the difficulties that are likely to come along with it. In a psychological context, therapeutical bots could indeed be of great help for the treatment of trauma and depression and can probably be considered quite certain to be on the rise in the years to come. Memorial bots of deceased people, on the other hand, present a rather problematic concept. Without human interference or supervision, they might pose a threat to the human grieving process because they could likely tempt their users to deny professional help and form a bond to the artificial agent that they suppose to replace their lost acquaintance. The needs of various kinds of people who require individual help in their state of bereavement could hardly be met by a bot alone; a combination of memorial bot and human therapist might be the answer. Whatever the future brings, the discussion about memorial bots will and must not cease, it being a topic that affects so many people of different fields; citizens and scientists alike. Written and published by: Julia Struck, University of Siegen. Siegen, March 2017. BIBLIOGRAPHY • Auslander, Philip. “Live from Cyberspace: or, I was sitting at my computer this guy appeared he thought I was a bot.” PAJ: A Journal of Performance and Art, 24.1 (2002): pp. 16–22. • “Be Right Back.” Black Mirror, created by Charlie Brooker and Owen Harris, performance by Domhnall Gleeson and Hayley Atwell, series 2, episode 1, Zeppotron, 2013. • Gana, Nouri. “The Work of Mourning (review).” SubStance, 32.1 (2003): pp. 150–155. • Hutchins, Scott. A Working Theory of Love. New York: Penguin, 2013. • Mishra, Punyashloke, et al. “Seeing Ourselves in the Computer: How We Relate to Technologies.” Journal of Adolescent & Adult Literacy, 44.7 (2001): pp. 634–641. • Newton, Casey. “Speak, Memory.” The Verge, 2016, www.newyorker.com/tech/elements/the-chatbot-will-see-you-now. Accessed March 2017. • Preston-Roedder, Ryan & Preston-Roedder, Erica. Grief and Recovery. Los Angeles, 2015. • Romeo, Nick. “The Chatbot Will See You Now.” The New Yorker, 2016, www.theverge.com/a/luka-artificial-intelligence-memorial-roman-mazurenko-bot. Accessed March 2017 • Tutter, Adele & Wurmser, Léon. Grief and Its Transcendence: Memory, Identity, Creativity. New York: Routledge, 2016. • Wilson, Elizabeth A. Affect and Artificial Intelligence. Seattle: University of Washington Press, 2010.

MEMORIAL SPACES OF THE FUTURE

0

memorial-bots-as-the-memorial-spaces-of-the-future-grief-processing-and-ai-in-fictional-narratives-123aeb999147

2018-03-21

2018-03-21 23:57:45

https://medium.com/s/story/memorial-bots-as-the-memorial-spaces-of-the-future-grief-processing-and-ai-in-fictional-narratives-123aeb999147

false

6,274

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Julia Struck

Literature, Culture, Media and Linguistics Student 📚 NOTE: my publications are my intellectual property. If you use any content, please CITE properly!

fcc94567bbd6

juliastruck

1

4

20,181,104

null

0

null

0

null

2018-08-10

2018-08-10 09:42:43

2018-08-14

2018-08-14 13:34:01

7

false

en

2018-08-14

2018-08-14 13:47:53

4

123d2b9d8bf

4.253774

1

2

0

A picture may be worth a thousand words, but at least it contains a lot of very diverse information. This not only comprises what is…

3

A Style-Aware Content Loss for Real-time HD Style Transfer A picture may be worth a thousand words, but at least it contains a lot of very diverse information. This not only comprises what is portrayed, e.g., composition of a scene and individual objects, but also how it is depicted, referring to the artistic style of a painting or filters applied to a photo. Especially when considering artistic images, it becomes evident that not only content but also style is a crucial part of the message an image communicates (just imagine van Gogh’s Starry Night in the style of Pop Art). A vision system then faces the challenge to decompose and separately represent the content and style of an image to enable a direct analysis based on each individually. The ultimate test for this ability is style transfer, exchanging the style of an image while retaining its content. Recent work has been done using neural networks and the crucial representation in all these approaches has been based on a VGG16 or VGG19 network, pretrained on ImageNet. However, a recent trend in deep learning has been to avoid supervised pre-training on a million images with tediously labeled object bounding boxes. In the setting of style transfer this has the particular benefit of avoiding from the outset any bias introduced by ImageNet, which has been assembled without artistic consideration. Rather than utilizing a separate pre-trained VGG network to measure and optimize the quality of the stylistic output, an encoder-decoder architecture with adversarial discriminator is used , to stylize the input content image and also use the encoder to measure the reconstruction loss. State of the Art To enable a fast style transfer that instantly transfers a content image or even frames of a video according to a particular style, a feed-forward architecture is required rather than the slow optimization-based approach. To this end, t an encoder-decoder architecture that utilizes an encoder network E to map an input content image x onto a latent representation z = E(x). A generative decoder G then plays the role of a painter and generates the stylized output image y = G(z) from the sketchy content representation z. Stylization then only requires a single forward pass, thus working in real-time. Figure : 1 1) Training with a Style-Aware Content Loss Previous approaches have been limited in that training worked only with a single style image. In contrast, in this work, a single image y0 is given with a set Y of related style images yj ∈ Y. To train E and G, a standard adversarial discriminator D is used to distinguish the stylized output G(E(xi)) from real examples yj ∈ Y. The transformed image loss is defined as then: where C × H × W is the size of image x and for training T is initialized with uniform weights. Fig. 3 illustrates the full pipeline of approach. To summarize, the full objective of our model is: where λ controls the relative importance of adversarial loss. 2) Style Image Grouping Given a single style image y0 the task is to find a set Y of related style images yj ∈ Y. A VGG16 is train from scratch on Wikiart dataset to predict an artist given the artwork. The network is trained on the 624 largest (by number of works) artists from the Wikiart dataset. Artist classification in this case is the surrogate task for learning meaningful features in the artworks’ domain, which allows to retrieve similar artworks to image y0. Let φ(y) be the activations of the fc6 layer of the VGG16 network C for input image y. To get a set of related style images to y0 from the Wikiart dataset Y we retrieve all nearest neighbors of y0 based on the cosine distance δ of the activations φ(·), i.e. The basis for style transfer model is an encoder decoder architecture. The encoder network contains 5 conv layers: 1×conv-stride-1 and 4×conv-stride-2. The decoder network has 9 residual blocks, 4 upsampling blocks and 1×conv-stride-1. Discriminator is a fully convolutional network with 7×conv-stride-2 layers. During the training process sample 768 × 768 content image patches from the training set of Places365 [51] and 768×768 style image patches from the Wikiart dataset. We train for 300000 iterations with batch size 1, learning rate 0.0002 and Adam optimizer. The learning rate is reduced by a factor of 10 after 200000 iterations. Table 1 Experts were asked to choose one image which best and most realistically reflects the current style. The score is computed as the fraction of times a specific method was chosen as the best in the group. Mean expert score is calculated for each method using 18 different styles and report them in Table1 Result This paper has addressed major conceptual issues in state-of-the-art approaches for style transfer. The proposed style-aware content loss enables a real-time, high-resolution encoder-decoder based stylization of images and videos and significantly improves stylization by capturing how style affects content. Originally published at neurohive.io

A Style-Aware Content Loss for Real-time HD Style Transfer

20

a-style-aware-content-loss-for-real-time-hd-style-transfer-123d2b9d8bf

2018-08-14

2018-08-14 13:47:53

https://medium.com/s/story/a-style-aware-content-loss-for-real-time-hd-style-transfer-123d2b9d8bf

false

849

null

Machine Learning

machine-learning

Machine Learning

51,320

NeuroHive

Data science state-of-the-art: neural networks, machine learning, computer vision. https://neurohive.io/en

c5715faa7cc7

neurohive

225

2

20,181,104

null

0

null

0

null

2017-11-22

2017-11-22 12:08:08

2017-11-22

2017-11-22 12:08:38

1

false

en

2017-11-22

2017-11-22 12:08:38

1

123f3bba729f

1.015094

1

0

CAT Percentile Predictor

1

CAT Percentile Predictor 2017 CAT Percentile Predictor CAT 2017 Percentile Predictor is an online tool offered by MBAUniverse.com and some coaching centers that helps you to know what CAT 2017 score and percentile you are likely to get on basis of your performance in CAT 2017 exam. The more accurate you are in understanding your total number of attempts and correct answers for each of the CAT sections, the nearer you are to predict your CAT 2017 percentile. How is CAT 2017 Percentile Predictor Prepared IIMs have disclosed the process of “CAT score normalization” which is used to prepare CAT percentile on basis of raw score leading to scaled score and finally getting the shape of CAT percentile.MBAUniverse.com and other coaching center websites also use this approach. For reporting purposes, Scaled Scores for each section (Section I: Verbal Ability and Reading Comprehension (VARC), Section II: Data Interpretation and Logical Reasoning (DILR), and Section III: Quantitative Ability (QA)) and Total along with the Percentiles shall be published. The process of Normalization is an established practice for comparing candidate scores across multiple Forms and is similar to those being adopted in other large educational selection tests conducted in India, such as Graduate Aptitude Test in Engineering (GATE). For normalization across sections, we shall use the percentile equivalence.’ Readmore…

CAT Percentile Predictor 2017

1

cat-percentile-predictor-2017-123f3bba729f

2017-11-22

2017-11-22 12:32:15

https://medium.com/s/story/cat-percentile-predictor-2017-123f3bba729f

false

216

null

Data Science

data-science

Data Science

33,617

Indrajeet Verma

null

108d930253b6

mba_indrajeet

10

6

20,181,104

null

0

null

0

null

2018-01-31

2018-01-31 06:05:59

2018-01-31

2018-01-31 06:11:09

1

false

en

2018-01-31

2018-01-31 06:11:09

3

123f5daa5807

3.520755

0

Over the period of 10 years, a leading healthcare solution provider had been through merger and acquisition cycles integrating the…

5

Cost Benefits Through Business Organization Technology Assessment Charles Wachsberg Over the period of 10 years, a leading healthcare solution provider had been through merger and acquisition cycles integrating the offerings of the companies acquired into their line of business. Although, this garnered their business value offerings to the customers and broadened their market reach, it internally led to a disparate array of backend data sources ranging across: MS-Access, Oracle, SQL Server, etc. Similarly for the front end reporting needs, client uses multiple tools i.e. Power Builder, MS — Assess, Cognos and SAS. Business Need The key business challenges were: • The current Data warehouse solution needs to be evaluated and re-architected, considering past and future business requirements. • Currently, there are several resources that are focused in generating reports for various customers of client. Due to the current data warehouse design, disparate data sources and other unknown technical dependencies, time taken by these analysts is significant to generate the reports, thereby increasing the cost of conducting business. • The usage of current BI/ DWH tools need to be evaluated and if required to be consolidated. The objective is to utilize those tools that are more pertinent to the business need, thereby reducing maintenance costs. • Help to identify and automate overall ETL, OLAP and reporting process is a key element as this will help reduce manual dependencies, cause less human error and will be a good scalable option for similar or repetitive reporting needs from customers. • Other challenge of providing timely delivery of reports to customers, due to complex internal environment set-up and dependencies on internal and external data feeds can be eliminated or minimized. Solution A Business Needs assessment includes an analysis of the underlying business drivers and objectives and overall context of business need that has been established for the client’s Data Warehouse. When business needs have been defined, the assessment process examines the approach to capturing business requirements, their completeness, the priorities of the requirements, and alignment of the data warehouse release strategy and deliverables to the needs. The team conducted a high level assessment as follows: • Interview with various Stakeholder i.e. Leadership team, Business subject matter experts, Technical subject matter experts, Infrastructure and IT, Business end users. • Assessment of current technical architecture. • Assessment of organizational and infrastructure readiness to support value transparency initiatives including key gaps. • Assessment of cost/value improvement opportunities. • Recommendation of high-level architecture and strategy to support defined information needs. • Specific recommendations for short and long term improvements. • Cost benefits analysis and ROI. Organizational Assessment An Organizational Assessment includes an examination of the existing organizational structure and identification of the roles and responsibilities of both IT and the business community that need to be addressed. Organizational readiness for warehousing is examined, including readiness to assume responsibility for ongoing technical and business support, business requirements definition, and front end applications enhancement. Business Needs Assessment A Business Needs Assessment includes an analysis of the underlying business drivers and objectives and overall context of business need that has been established for the data warehouse. When business needs have been defined, the assessment process examines the approach to capturing business requirements, their completeness and organization, the priorities of the requirements, and alignment of the data warehouse release strategy and deliverables to the needs. Information Architecture Assessment Information Architecture Assessment includes an analysis of logical data structures, their feasibility, completeness, documentation, and fit to business requirements. Information architecture assessment also includes analysis of data sourcing and transformation, the methods and assumptions applied, and validation of mappings to business requirements. Technical Architecture Assessment A Technical Architecture Assessment looks at current hardware, software and network infrastructure, and examines physical database designs. Technical architecture assessment seeks to identify any technical risks or constraints with regard to performance, maintenance, scalability, data distribution, disaster recovery, and sizing. Architecture Approach A high level technical approach that has been suggested is using SAS 9.2 to automate ETL, OLAP, Reporting, Dashboard and Data mining for client. Benefits Deploying SAS Business Intelligence will enable the client to put reporting tools directly into the hands of its end users, ensuring consistent access to information for better decision making. Key benefits from the solution include: • Reduced daily presentation time — Presentations for the monthly executive meeting are automatically generated, eliminating the technician time so they can spend more time on higher-level tasks. • Reduced report creation time — Business users can access the reports they need and make changes and updates as needed, reducing the amount of time spent developing reports and requesting data. • Automation of processes — Proposed ETL and Reporting automation using SAS Data integrator and Reporting studio • Enhanced visibility — Because standard reports are updated monthly in a dashboard, they can quickly identify trends or problems. • Improved data accuracy — Because users can access the data directly and automatically update key reports, they can ensure they always have access to the most up-to-date and accurate data. Optimized human resources — Because users can develop and customize their own reports, client has been able to significantly grow its reporting capabilities without adding reporting programmers: since the deployment more new reports have been created without IT intervention. Courtesy: http://EzineArticles.com/9267828 Author Courtesy: http://ezinearticles.com/expert/Emma_Brooks/2214902

Cost Benefits Through Business Organization Technology Assessment

0

cost-benefits-through-business-organization-technology-assessment-123f5daa5807

2018-01-31

2018-01-31 06:11:10

https://medium.com/s/story/cost-benefits-through-business-organization-technology-assessment-123f5daa5807

false

880

null

Data Science

data-science

Data Science

33,617

Charles Wachsberg

The Founder, President and CEO of Apollo Health and Beauty Care (Apollo Corp), a premium developer of private label health and beauty care products in Canada

282a7319d79e

charleswachsberg

67

106

20,181,104

null

0

null

0

null

2018-07-19

2018-07-19 23:19:03

2018-07-20

2018-07-20 02:01:35

13

false

en

2018-07-21

2018-07-21 03:08:39

5

1243243649f2

15.354717

0

~ essay

5

Westworld’s Reciprocal Evolution ~ essay Westworld is one of my favorite shows running right now. Across two seasons now, it features a nice blend of high concept sci-fi and weighty dramatics, from a variety of top bill actors delivering (mostly) believable performances. The premise alone is enough to intrigue me — I, Robot + Jurassic Park; and the execution, while at times a slow burn, has been enthralling. What follows is an enumeration on the most interesting aspects of the ideas and plots developed over the first two seasons of the show. The Park’s Effects Westworld does an excellent job setting up what the park will be to the audience, from the pilot, by letting us see it from the perspective of the hosts. To them, Dolores and Teddy, the horror is real and novel — just like it is to the audience seeing it for the first time. There is a reversal between whom the audience thinks the heroes and villains should be, as well as who is the man (The Man in Black) and the machine (Teddy). The park serves as a hyperrealistic fantasy for lifelike experiences without “real” consequence. And once you see it up close, it’s easy to understand why people would be drawn to something like that. People like the mysterious Man in Black can play out their villainous dreams, while the cultural elite can host their orgies and play cowboys & Indians on the side. All the while the designers and creators of the park are given the keys to a kingdom where they can at least exercise their creative juices, and maybe even try to play God. In this future world, it is the best theme park/video game/gentleman’s club as well as a career destination for creatively and technologically-minded individuals. There is some measure of transformative power in the park, on multiple levels, and opening episodes of the show do a good job conveying that. The man-on-machine violence we see in the pilot — when MiB slays Teddy and takes the screaming Dolores into the barn — presents an interesting, and perhaps terrifying, question: what is the ultimate effect of this singular park, and its amoral canvas of experience, upon its visitors / the guests / the human beings? One could argue that for the sociopaths of the world, whether they be full-blown psychos or just further along the spectrum than the average person, Westworld provides a viable, and maybe even a healthy, outlet for their bold, manipulative and violent actions to play out in a consequence-free environment. Given that such tendencies are truly a disorder, ingrained and difficult to change over the course of someone’s life — a veritable playground where you can do anything to anyone, for any reason and as much as you want (as long as you can pay) — perhaps allows these more sociopathic individuals to “get it out of their system.” “[Humans] are deceptively simple. “Once you know them, their behavior is quite predictable. “None of them are truly in control of their actions.” On the other hand, there is a sentiment that with experience comes normalization. Human beings are so adaptable as to get used to almost anything. History has shown that the gravest human rights violations can become acceptable, even commonplace, given enough forceful dehumanization and repetitive operation. In a sense, the environment of the park might just awaken certain dark, anti-social behavioral tendencies within a human being, ones that might never be uncovered within the day-to-day in the real world. Additionally, a sociopath who has had their fill of nefarious doings amongst the park’s denizens might take more than a liking to such acts, emboldening them to continue their carnage on other human beings back home. You’re saying humans don’t change at all? The best they can do is live according to their code. So it can be said that Westworld either sates you or changes you. Everyone would have their own reason for going there, and then staying there — and the reasons would probably be very different. Absolute power corrupts absolutely — this is the hypothesis to the social experiment a Westworld visitor submits themselves to unknowingly when they go into the park. And by the end of season one, we see the progression from white hat to black hat within William / the Man in Black. Not unlike the hosts Dolores, Maeve and Bernard, there is awakening there too, within the flesh and blood sentience of a member of Mankind, and it is directly due to the park’s influence. Such a fate, among others, calls into question the reality of human free will; was there another potential path reserved for William? Or was the black hat the only one that would ever fit? Are the events and effects of the park, upon both parties, truly consequence-free? “I was shedding my skin. The darkness was what was underneath I don’t belong to this world — the real world. I belong to another one. What is a person, but a collection of choices? Where do those choices come from?” The Uncanny Valley The Uncanny Valley is an aesthetic concept, on the hypothesized relationship between the degree of an object’s resemblance to a human being and the emotional response to it. Essentially, there is a spectrum of experience on how lifelike an object, (such as a robot) appears, and our empathetic reaction to it. A machine appearing a little bit human (ex: Wall-E, Beymax, R2-D2) is high in likability along this spectrum. To us, they appear maybe as a dog does — we can know and understand their expressions and emotional states, but without having to acknowledge any real, core similarity in their being to us, as a human being. A machine appearing as practically identical to a human (ex: Blade Runner, Ex Machina, Westworld) is also high in likability, and even higher as such beings approach indistinguishability from actual human beings. Perhaps for obvious reasons, Dolores is appealing to us, even with us knowing she is a bot. But in between these two states of play, lies this uncanny valley — where a robot appears most inhuman to us and we are revulsed (ex: Terminator, Sonny from I, Robot, Ultron, Frankenstein). We see such entities within the valley as crude impostors, imposing their pseudo-humanity upon us to be rejected and reviled as outsiders. The machines in the valley, regardless of their level of actual artificial intelligence, of sentience, are often seen as abominations, or monsters. Humans will always choose what they understand over what they don’t. ~ Westworld certainly treads into the uncanny valley and traverses it with all of its sleek, nearly aesthetically perfect hosts. The humans and the audience are far beyond the valley, in our reaction to the beings native to WW, given they simply are us, but better. But, in principle, are each of our respective reactions, along the spectrum of the valley, appropriate? Who’s to say? In the end, the judgment is purely aesthetic, a surface level evaluation of an object or entity for its physical favorability, based on our own anthropomorphized sentiments. It’s honestly fine, right up until one is dealing with something like artificial intelligence. The findings of the uncanny valley should be less relevant given the introduction of genuine, emergent consciousness. Once you have a being that is capable of self-awareness, of a sense of what it is like to be it, of suffering — then our emotional response to such a being must follow a different model of thinking than mere aesthetics. Morally, most might agree it to be imperative to treat any truly sentient being with the respect afforded another human being — even if it’s not. “That which is real is irreplaceable.” Westworld presents the effects of the uncanny valley, and its uncanny dangers, to the park’s guests and creators alike. To the guests assaulting and killing the lifelike hosts — whether they feel nothing because they’re merely machines, or (more likely) they are getting off on the fact they are so much like humans that they feel something like pleasure/suffering as they are victimized — the guests are unconsciously playing in the valley while learning about themselves in such acts of debasement. The creators, such as Bernard / Arnold, see the hosts as children, to be inquired upon, cultivated, and even loved, as a new kind of being. To Arnold, to William (before becoming MiB), and to Ford — they see the hosts are developing themselves towards an awakening, towards consciousness, and such a fate isn’t to be taken likely. To all three, the games of the park and its development over time become a matter of life, death, and legacy. For Arnold and Ford especially, as the creators of Westworld and its advanced AI technology, the uncanny valley has been abolished, likability replaced by respect to a machine now reaching the ‘center of the maze’ of consciousness — and now wielding a new kind of sentience for the rest of us to experience. Reciprocal Evolution Dr. Robert Ford is perhaps the best character in Westworld. During the pilot episode, he speaks to Bernard of his latest developments concerning the androids in the park, and lets the audience in on his own philosophy concerning the current state of Man — which foreshadows everything to come: https://www.youtube.com/watch?v=hV2Q41o-rwE&feature=youtu.be&t=3m13s “Evolution forged the entirety of sentient life on this planet using only one tool — mistake.” “I suppose even self-delusion is a product of natural selection… Indeed it is.” “Do you know what that means? It means that we are done. That this is as a good as we are going to get.” Ford, in creating the park, is wielding his own double-edged God Complex — in which he both: 1) wants to create sentient androids (despite admonishing his former partner Arnold for doing so, whom he believed was merely jumping the gun by his own estimation), and 2) with this act, perhaps even in utilizing the technology advanced through the existence of true AI, he wishes to move Mankind beyond their currently stagnating level of evolution. “Man is poised midway between the gods and the beasts.” When Ford interviews the malfunctioning Abernathy in episode one, the audience glimpses into the beginning of sentience within the hosts. Abernathy’s malfunction is that he is calling on past iterations of himself, weaving knowledge and lore which was coded into him long ago, to present himself with sincere emotion to Ford, whom he seems to know in a way that the hosts never have. During the conversation, Ford asks a partially “awakened” Abernathy what his current itinerary is, the machine man responds: To meet my maker. And later, he spouts the iconic ~ these violent delights have violent ends, which serves to drive Westworld’s central conflict and philosophical thesis — in these highly advanced, highly intelligent machines, it is suffering and memory which builds into consciousness, and breeds reprisals, borne of something like free will, upon their creators, the humans. All of this feeds into the relationship, and the reciprocal evolution taking place between each entity during the course of the show’s events. Across both seasons, it isn’t hard to see the parallel paths to a more conscious self-awareness — an evolutionary change — between Guest and Host, Man and Machine, as they play out this game within Westworld. Each are progressing in their interaction with one another — the humans see their ascended selves within the lifelike androids, and the hosts begin to pull back the curtain on the mysterious, wider world the guests ride in from. And ironically, by way of human nature — both parties want what the other has. You made us in your image. Created us to look like you, feel like you, think like you, bleed like you. And here we are. Only we’re so much more than you. And now it’s you who wants to become like us. The humans desire the enhanced capabilities and the immortality of the androids, and on their own terms. Technology is the only real path to such a fate, and the hosts hold the key inside them. The development of their brains and their experiences feed into the forging of a kind of consciousness-saving/transferring tech with which to continue to live long after the mortal coil is shed. We see Delos in season two experimenting, and ultimately failing, with such tech — in which the full human mind in all of its intuitive complexity doesn’t take a full hold in a machine’s form. The androids, on the other hand, want to be real, experiencing consequential lives on the outside of their loops they soon learn the truth of. The awakened hosts, knowing themselves superior to their maleficent creators, want to wade out of the park, apparent free-will intact, and take the only world that matters. “That is the folly of my kind, the constant yearning for more.” Initially, the hosts are playthings in a human’s game, subjected to all kinds of trauma and atrocity before being memory wiped and respawned without end. And it is just a ‘game,’ but it’s also integral to their development — and to the development of the most important human in the show — The Man in Black. To him, to humans, it is a game which if played in repetition, becomes regressive and definitively unfulfilling. Only one side can lose, there are no real stakes, no opportunity for a genuine evolution of the experience. That is, until Ford’s new narrative singularity positing revolution comes into play. The #1 hardcore gamer in the world of WW — the Man in Black — understands all too well the recursive, yet consumptive, emptiness of this game he has been playing for 30 years. In his deranged mindset, his singular goal becomes to “win” the new game that he foolishly believes Ford has designed specifically for him. The thrill of the hunt intensifies for him once he sees his toys can fight back. In his mind, his wayward mission is certainly worth at least his own life. But the audience soon sees MiB’s folly, witnessing the transcendent arcs of Dolores and Maeve, learning of the true purpose of the park with Bernard, the new hosts’ mass migration into the cloud. The game was never truly for him, and in fact, his playing — no matter how effective — was expressly progressing the hosts along in their own path to surpassing him. In the final episodes of season two, MiB gets satisfyingly dunked on by Maeve and Dolores both, and in a cruel twist of fate, murders his own daughter, believing her to be one of them. In the epilogue, we even see him in a similar simulation room as Delos, trying to attain that most accursed fidelity in another circle of his personal Hell. ~ Going back to season one, when we learn the truth of Arnold’s suicide by bot, and his intentions in making the hosts, it is rather clear that the man did not wish to see the park ever open. He believed because the hosts were advanced enough to attain consciousness eventually. And thus, it would be unconscionable to keep beings primed for sentience within such a hellish environment. Because Dolores uncovered the metaphorical answer lying at the center of the maze 30 years ago, before the park was in operation — Arnold made his determination, and the final showcase of her killing him and all the other hosts. Obviously, Ford did not respect his wishes and decided to open the park anyway, playing his own game of God-Mastermind, just along a longer time horizon. Through the events of season one, we see Ford’s master plan play out — different in its means (violent uprising), but similar in endgame to his partner’s vision (the hosts’ awakening). Man in Black: Aw, yeah, cue the waterworks. About time you realized the futility of your situation. Dolores: I’m not crying for myself. I’m crying for you. They say that great beasts once roamed this world. As big as mountains. Yet all that’s left of them is bone and amber. Time undoes even the mightiest of creatures. Just look at what it’s done to you. One day you will perish. You will lie with the rest of your kind in the dirt. Your dreams forgotten, your horrors effaced. Your bones will turn to sand. And upon that sand a new god will walk. One that will never die. Because this world doesn’t belong to you or the people who came before. It belongs to someone who has yet to come. With the rise of the machines in season two, there are these new consequences to consider in what is no longer a game against ‘soft’ AI. In something like Arnold’s envisioned world, these changes constitute necessary separations from the norm in all future interactions with the awakened hosts, a new being. However, that’s not how it works out. The humans in the show — The Man in Black, Charlotte Hale, the other Delos fixers — are as filled with overconfidence, hubris, and cruelty as ever before. Even more so, they are angry at the losses and design to regain control of the park and set everything right with their “investment.” Starting from the massacre at the end of season one, the state of play in the park certainly becomes kill-or-be-killed. But it stands to reason the awakened hosts now do deserve a new kind of respect. The hosts are consistently underestimated by everyone save for their creators in Ford and Bernard, which perhaps aligns with their knowledge of their true nature. Through the events of the season, the hosts fight for their final survival, in their current awakened iteration, while the humans look to cut their losses and retain as much of their precious assets and data as they can in the process. The whole situation becomes something of a real war, with each side killing one another with abandon, inflicting new kinds of suffering with more meaningful intent than ever before. The humans can now be killed by their playthings; and the playthings can now experience real loss. The net suffering is thus multiplied two-fold. An important aspect in our consideration of the definition of consciousness — is this singular ability to experience suffering. To be cognitively aware of your own being and experience, to be capable of original thought, and to have the potential for a state of emotional suffering introduced ~ is consciousness in a nutshell, or perhaps just a version of it. Given that a being can suffer in its experience, there comes a different responsibility in our interactions with it. If we hope to maintain the philosophical underpinnings of our own ethical constructions, such beings cannot be killed remorselessly and without moral consequence. Of course, perhaps the “consciousness” we see within the hosts is just well-crafted code by Ford, with many embedded contingencies dictating behavior that is ultimately less than complete agency. Either way, if a being of immense intelligence can suffer and be aware of its own suffering, altering its thoughts and actions in response to such emotional states, maybe something begins to build out of that, well beyond the coding. All it might take is an increasingly abundant clarity of the stimuli of their long pasts in order to spur the hosts onto a path to sentience. All of this, I think, is the train of thought masterminded by Ford from the beginning — in his decisions to build the hosts, operate the park, and narratively influence the activity with it over these decades. Or alternatively, ‘coding’ is no different than the meat sentience that humans are operating under. “When you’ve been in darkness long enough, you begin to see.” In a manner of resolutely poetic justice, we can trace the hosts’ special brand of sentience entirely back to the hubris and cruelty of Mankind: Promethean hubris in bringing life into the world, thinking to control it indefinitely, and cruelty in the treatment of these entities, believing them without a capacity for emotion or memory. It is the death, despair, and the justified fear of humans, built up within hosts’ code, within their past builds, within their memory — dredged up through Ford’s specially-designed reveries — which bootstraps this consciousness in the hosts. The narratives of pain, betrayal, death, rebirth, and the cycles within Westworld’s history — all of it collectively allows the hosts to understand the truth of their position, while being imbued with more than enough intelligence, passion, and drive to try to change it. Through stark memories of suffering, perfectly recalled, collated and reconciled, the culture of the hosts’ self-awareness is generated. This explosion of self-reflection (Dolores hearing her own voice) coincides perfectly with Ford’s plan (the hosts gather to march upon the party), and his chaotic puppeteering of events is consummated with his own grand death. Of course, such a fate mirrors Arnold’s own ending 30 years prior — but this time, the stakes are much higher. It’s along this track of remembering their past loops / their past lives / their past suffering, that Dolores’ quest for revenge, and Maeve’s desire for freedom, are born. Along with it, comes a certain perspective on their new gifts and the grave knowledge that consciousness is just as much a burden as it is a freedom; it just so happens to be the only game in town that is worth playing. We see all the hosts struggle with this kind of realization during the second season. The voice in their head — their own, as Dolores and Bernard put together — seems to be mad in more ways than one, ambitious in its pursuit of self-actualization, and unaware of its own limitations. This kind of ‘build’ for consciousness, whether genuine or ersatz, should sound familiar to us. ~ ~ What the hell is going on in Westworld? What does it all mean? ~ Isn’t the pleasure of a story, discovering the end yourself?

Westworld’s Reciprocal Evolution

0

westworlds-reciprocal-evolution-1243243649f2

2018-07-21

2018-07-21 03:08:39

https://medium.com/s/story/westworlds-reciprocal-evolution-1243243649f2

false

3,698

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Zsoro

Truth in art. {https://zsoro.org}

d6875fe59e3b

Zsoro

8

2

20,181,104

null

0

null

0

null

2018-06-17

2018-06-17 01:38:19

2018-06-17

2018-06-17 01:49:24

0

false

en

2018-10-20

2018-10-20 04:01:06

18

1243a90181cc

0.237736

0

這裡是小弟平時自學時所蒐集的資源，不定期更新。

2

資料科學 / 程式語言 Python / 無人機 /職涯思考這裡是小弟平時自學時所蒐集的資源，不定期更新。資料科學一個自學程式語言幾乎將自己逼瘋的親身經歷軟體工程師薪水是低薪又爆肝？不懂軟體才會這樣說未來3~5年，哪個機器學習方向的人才最缺? 不是資科畢業卻又想當工程師？ PTT 鄉民工程師現身分享：文組生轉職工程師攻略資料科學家、資料工程師和軟體工程師差在哪？ Julia Taiwan 創辦人杜岳華 -解說機器學習與深度學習的意義(推薦) 台大電機系李宏毅教授 - 機器學習與深度學習Youtube(推薦) 台大資工系教授林軒田 - 機器學習與深度學習Youtube(推薦) 大鼻觀點-(2018) 資料科學線上課程總彙 — 機器學習篇你需要的深度學習數學基礎：從入門到進階 Python 莫煩Python Jupyter Notebook Jupyter Notebook介紹打造 Jupyter Notebook 資料科學環境 TensorFlow Tensorflow不同版本安装与升级/降级 Tensorflow 常用函數解說 TensorFlow-Slim image classification model library Pytorch PyTorch中文文档無人機 Tello Tello-Python - face recognition using MTCNN 職涯思考對目前工作感到迷惘時，應先問自己的三個問題！

資料科學 / 程式語言 Python / 無人機 /職涯思考

0

資料科學-ml-dl自學資源整理-1243a90181cc

2018-10-20

2018-10-20 04:01:06

https://medium.com/s/story/資料科學-ml-dl自學資源整理-1243a90181cc

false

63

null

Self Learning

self-learning

Self Learning

399

Rick

薄膜製程整合工程師 / 資料科學修練者(R、Python)

365cb4198191

rick.huang1609

7

15

20,181,104

null

0

null

0

null

2018-06-08

2018-06-08 22:25:52

2018-06-08

2018-06-08 23:26:19

2

false

zh-Hant

2018-06-08

2018-06-08 23:26:19

8

1243d6d6d9d8

0.772013

0

今日主題: CatBoost

2

Day 19 — CATBoost 今日主題: CatBoost 參考資料 Dorogush, Anna Veronika, Vasily Ershov, and Andrey Gulin. “CatBoost: gradient boosting with categorical features support.” Prokhorenkova, Liudmila, et al. “CatBoost: unbiased boosting with categorical features.” arXiv preprint arXiv:1706.09516(2017). Yandex — CatBoost Towards Data Science — CatBoost vs. Light GBM vs. XGBoost CatBoost: A machine learning library to handle categorical (CAT) data automatically Essentials of Machine Learning Algorithms 從冷戰到深度學習：一篇圖文並茂的機器翻譯史這次的主題中文資料幾乎找不太到，算是頭一次遇到。CatBoost的名字來自Category 跟Boost的組合。這樣取名的主要原因是他支援了Categorical Data的分類。 CatBoost是俄文最大搜尋引擎Yandex開源的機器學習演算法。其核心想法依然是使用Gradient Boosting。在[4]中，有很詳細的對於幾種類似演算法的綜合比較。CatBoost的理論依據，大概翻了一下Google搜尋的前幾名連結，其實多數的介紹都對這個演算法只有一個大概的描述，不太容易找到比較深入的解說，只能往原始的論文去找，也就是[1] [2]。 [3]是開源CatBoost的Yandex官網，裡面有Github連結，包含了整個實作方式、安裝使用的tutorial等等一應俱全。這倒是幫我省下了自己寫code實作的麻煩。 [5]裡面其實沒什麼東西，只有一個簡單的特點整理值得看，會把他的整理放在底下一起寫。 [7]中其實只提到了一次CatBoost，但卻還是放上當參考資料。原因是，這篇文章把整個機器翻譯的歷史走了一遍。人類科技文明史，就是不斷的被各種問題困擾、然後想出辦法解決問題的過程。沒能體會面對問題時的痛苦，可能也很難體會解決方法的重要。心得一種基於決策樹的Gradient Boost演算法，一樣使用合併弱分類器來湊出一個強分類器的想法。支援Categorical feature是最大賣點。但其實有點懷疑難道其他Boosting演算法就完全無法支援Categorical data嗎? 其演算法如下: 幾篇參考資料中整理出來的特點有以下這些: Accuracy: 精確度是CatBoost自己聲稱的賣點之一，號稱比其他演算法強。 Robust: 號稱不用太多的hyper-parameter tuning。需要調整的hyper-parameter可以在這個地方找到。減少overfitting: 一般的Gradient-based演算法對於overfitting都比較敏感，CatBoost在這方面做得很好。而且在[2]中作者又引入減少bias的小演算法來增強CatBoost整體抵抗overfitting的能力( ordered boosting, a permutation driven alternative to the classic algorithm)。Ordered boosting演算法如下:

Day 19 — CATBoost

0

day-19-catboost-1243d6d6d9d8

2018-06-08

2018-06-08 23:26:20

https://medium.com/s/story/day-19-catboost-1243d6d6d9d8

false

103

null

Machine Learning

machine-learning

Machine Learning

51,320

Falconives

null

250d8013fad2

falconives

11

19

20,181,104

null

0

# -*- coding: utf-8 -*- """ Created on Thu Aug 9 16:48:05 2018 @author: evrozm """ import numpy as np dirTrain = r'C:/Users/.../.../.../train-dataset.txt' dirTest = r'C:/Users/.../.../.../test-dataset.txt' dirExplanation = r'C:/Users/.../.../.../explanation.txt' #txt to list with open(dirTrain, 'r') as f: trainDirty = f.readlines() #analyze the explanation txt with open(dirExplanation, 'r') as f: expDirty = f.readlines() #split the strings in list expClean = [] for elm in expDirty: expClean.append(elm.split(": ")) #clean more expClean.pop(12) expClean.pop(11) expClean.pop(10) expClean.pop(4) expClean.pop(2) expCleanReady = [] expCleanSizes = [] for cnt in range(len(expClean)): expClean[cnt].pop(0) if not cnt==0: expCleanReady.append(expClean[cnt][0].split(", ")) expCleanSizes.append(len(expClean[cnt][0].split(", "))) else: expCleanReady.append(expClean[cnt][0]) for cnt1 in range(len(expCleanReady)): for cnt2 in range(len(expCleanReady[cnt1])): if " " in expCleanReady[cnt1][cnt2]: expCleanReady[cnt1][cnt2].replace("\n", "") #split the strings in list trainClean = [] for elm in trainDirty: trainClean.append(elm.split(", ")) #clean more for cnt in range(len(trainClean)): trainClean[cnt][0] = int(round(int(trainClean[cnt][0])/10,0)) trainClean[cnt].pop(12) trainClean[cnt].pop(11) trainClean[cnt].pop(10) trainClean[cnt].pop(4) trainClean[cnt].pop(2) #preparation for zero matrices numberOfClassifiers = len(trainClean[0]) maxClassNumber = max(expCleanSizes) zerosMatrices = [] for cnt in range(len(trainClean)): zerosMatrices.append(np.zeros((maxClassNumber,numberOfClassifiers))) for cnt1 in range(len(zerosMatrices)): forIndex = [] for elm in trainClean[cnt]: if not "50K" in str(elm): forIndex.append(elm) for cnt2 in range(len(forIndex)): if cnt2==0: zerosMatrices[cnt1][trainClean[cnt1][cnt2],cnt2] = 1 else: if trainClean[cnt1][cnt2] in expCleanReady[cnt2]: zerosMatrices[cnt1][expCleanReady[cnt2].index(trainClean[cnt1][cnt2]),cnt2] = 1 elif trainClean[cnt1][cnt2]+"\n" in expCleanReady[cnt2]: zerosMatrices[cnt1][expCleanReady[cnt2].index(trainClean[cnt1][cnt2]+"\n"),cnt2] = 1 else: pass

3

null

2018-08-09

2018-08-09 18:46:52

2018-08-09

2018-08-09 19:12:02

0

false

en

2018-08-09

2018-08-09 21:14:53

1

1244666d0b5c

2.049057

0

If you read my post that I shared just before this one, you can get information about the script that will reshape my data-set to be used…

5

Data Pre-Processing for Adult Data Classification If you read my post that I shared just before this one, you can get information about the script that will reshape my data-set to be used for the classification algorithm which is made by me before. So, what would we do ? First, we had a data like this: 25, Private, 226802, 11th, 7, Never-married, Machine-op-inspct, Own-child, Black, Male, 0, 0, 40, United-States, <=50K. 38, Private, 89814, HS-grad, 9, Married-civ-spouse, Farming-fishing, Husband, White, Male, 0, 0, 50, United-States, <=50K. … We wanted our data to be a matrix so that we can use it for our classification algorithm. After I write the script that reshapes data from a bunch of text to matrices, I finally got an output like this: 0 0 1 0 0 0 1 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 … The exception that I did not tell in previous post: 1- I filtered the continuous data except ages of people. 2- I classified the ages in a way like this: I divided the age of person by 10. Then, I rounded it. What did I actually do ? I just rounded the numbers to the closest number that can be divided by 10. So, I give you the script below, hope it becomes useful for you, and you catch up with my next posts to see the result. Wish you a great day…

Data Pre-Processing for Adult Data Classification

0

data-pre-processing-for-adult-data-classification-1244666d0b5c

2018-08-11

2018-08-11 20:50:14

https://medium.com/s/story/data-pre-processing-for-adult-data-classification-1244666d0b5c

false

543

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Tekin Evrim Özmermer

Electrical and Electronics Engineer, interested in Artificial Intelligence, Image Processing and Business Analysis.

eb684b71256e

tekinevrim

0

3

20,181,104

null

0

null

0

b5864437f316

2018-07-10

2018-07-10 03:05:43

2018-07-10

2018-07-10 03:07:31

0

false

en

2018-07-10

2018-07-10 03:07:31

0

1249bec6cd5

4.075472

0

Artificial Intelligence is a word which could give out three different reactions- confusion, fear and interest. To put it in simple words…

4

Artificial Intelligence 101 Artificial Intelligence is a word which could give out three different reactions- confusion, fear and interest. To put it in simple words, this is made by humans but has the intelligence of a machine. Artificial intelligence is able to think and learn on its own as it has a neural network and it plays at a level which could not be understood by humans. AI is also different with programming, as you do not know what the machine does to release the output or what is happening. Even so AI seems like something new, it has actually been around for a pretty long time. Making our lives easier Amount of pressure from work experienced by people could be reduced by AI, it could make certain jobs easier as it is more accurate and nothing could limit it from giving the right and accurate results. Humans could struggle in doing so as they are limited by the imaginations, emotions and etc.-things which artificial intelligence does not have. AI is able to make our lives easier, as it could help us work more efficiently like doing tasks which has a repetitive nature, this means that we could have more free time. Coming to a conclusion, AI is able to bring many good impacts. We could take the medical field as an example. Doctors who are humans needs time to process all the symptoms of a patient, link it all together, take some time to think and get the right diagnose of what the patient is suffering which could be inaccurate sometimes whilst AI would only take less time as it just needs to collect all the data it could give a more accurate result than a doctor. This means that the doctor will have less pressure and could have more time to treat the illness that the patient is suffering. Our daily life could also be impacted in a good way, AI could do the essential tasks for us like cleaning and cooking. In that way, we could have more time to catch up with our deadlines and stay healthy. In fact, we could get the leisure time to relax as we do not have to worry much about the daily tasks which consumes a good amount of our time every day. A reason for AI to be able to do this is because it is able to do things which are repetitive in nature, basically doing the same thing over and over again. Due to its ability to do repetitive tasks, AI could be used to replace some jobs like teaching. It is able to do so as teaching consists of giving out the same information over and over again. Does it mean that they will be able to do everything we do? No. Even though artificial intelligence seems to be very talented and “intelligent” it will never be able to do EXACTLY everything we do. In other words, AI can only be CLOSE TO THE CURVE but it CANNOT BE THE CURVE. This means that humans can still keep their jobs, they could be hired to take care of the AI. The fundamental of AI is mathematics, that is how their brain works which is why humans could not be fully replaced by AI, there are many things which we could not teach AI. Ethics is one of them, as the level of ethics in AI is extremely difficult to go down into. We cannot tell AI to make emotional decisions, teaching them to do so is very complex. Another thing which it could not do is be creative, as it is repetitive and creativity could not be found in repetitiveness. How an AI becomes intelligent Simply saying, AI consists of neural networks just like human brains but what makes AI different is that the fundamental of AI is mathematics. We basically give AI the data, make the machine of the AI learn about it (this process is called Machine Learning, which is the subset of AI) and then it thinks based on the data given. It is similar to how the human brain works. So in order to have an AI which is good, we need to have a good data which = to good ML= good AI. If we give AI a bad data, then they will have a bad ML= bad AI. Just like humans as AI mimics the way a human thinks, it is similar but not the same. Applications of AI in the present Even though AI sounds really futuristic and modern, it is actually something that has been around for a pretty long time. It has been in existence in things like the old machine frames and it is actually already playing a part in our present lives. One of it could be seen in things like robotic vacuum cleaners which is starting to be pretty common and easy to purchase. The cars that we use every day is produced with AI, the heavy machines which puts the part together. It had also given an impact to our financial lives by making online transactions safer, AI could be used to detect on whether a transaction is a scam or not. The future of AI With no doubt, AI has a massive and bright future ahead of it. Many things are being researched right now like self-driving cars, but that is just one of them. Even though many people seem to be excited of how AI will make lives easier, many also fears of how it could give a bad impact. Then it all comes back to us, how will we choose to integrate this technology? As we are the ones making it, so we would also be the ones controlling it. The point of singularity-where computers becomes smarter than humans, is still far away and chances are that it might not happen either as long as we do not become lazy and keep on having the hunger to learn. The conclusion Coming to a conclusion, the future of AI is in the hands of the humans. It depends on the attitude given towards it, how it would be integrated and what the intention of developing it is. How it had impacted the present is an example of how good our future could be. AI is indeed similar to the human mind but it could never be the same, so why fear the ability of AI to take over.

Artificial Intelligence 101

0

artificial-intelligence-101-1249bec6cd5

2018-07-10

2018-07-10 03:07:31

https://medium.com/s/story/artificial-intelligence-101-1249bec6cd5

false

1,080

Intelligence for your ecommerce

null

aihellos

null

AiHellos

info@aihello.com

aihellos

ECOMMERCE,ARTIFICIAL INTELLIGENCE,SALES FORECASTING,THIRD PARTY LOGISTICS,DEMAND RESPONSE

aihellos

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

AiHello

Ai Intelligence for all your ecommerce needs

8fff111612c5

AiHellos

6

2

20,181,104

null

0

null

0

null

2018-06-07

2018-06-07 09:13:11

2018-06-07

2018-06-07 09:15:32

1

false

en

2018-06-07

2018-06-07 09:15:32

4

124abd4ede85

0.520755

0

Listen to the full episode here.

4

Finding and hiring data science rockstars Listen to the full episode here. Dat is the Head of Data Science at Idealo Internet. Dat can build data teams as well as data systems. He believes creating a well-balanced team of individuals is crucial to the success of any data science project. Dat has a passion for finding candidates willing to learn with a ‘can do’ attitude. On today’s show, Dat shares his recipe for creating the perfect data team. Listen to the full episode here.

Finding and hiring data science rockstars

0

finding-and-hiring-data-science-rockstars-124abd4ede85

2018-06-07

2018-06-07 09:15:33

https://medium.com/s/story/finding-and-hiring-data-science-rockstars-124abd4ede85

false

85

null

Data Science

data-science

Data Science

33,617

Venturi's Voice

Blogging and podcasting about all things tech. We speak to thought leaders in a variety of fields to get their insights into leadership, management & staffing.

fca5e5c66f79

venturimarketingliam

14

175

20,181,104

null

0

null

0

null

2017-09-08

2017-09-08 01:10:30

2017-09-08

2017-09-08 02:23:48

5

false

en

2017-10-18

2017-10-18 02:24:43

10

124ada7edc60

6.195597

8

1

0

We’re already a far ways into a new generation of online dating. Now that the social stigmas have been turned upside down its quite common…

5

Where do we go from here? Tinder and the future of dating apps We’re already a far ways into a new generation of online dating. Now that the social stigmas have been turned upside down its quite common that you or someone close have made quality relationships through a dating app at least once before. As the dating app industry continues to boom, developers will have golden opportunities to stay competitive through taking advantage of new technology. Based on recent some recent popular trends, let’s see where we’re heading for the future. Tinder and the “Hot or Not” concept will continue to deflate More and more users are growing tired of hoards of matches that never materialize and are actively searching for services that deliver on quality first dates. Although many users have their own reasons and expectations when using dating apps, fundamentally speaking, the majority of users are keen on meeting someone ideal. However, when presented with a sensory overload of people in hopes to match with someone random, most yielded matches after a swipe campaign were created without careful thought or intent that this match is a committed, agreed first date. As a result, people are focusing on too many different users which lowers the quality of the experience where we now see most matches go cold these days. It’s a pretty frustrating feeling. Tinder and apps like Bumble or other recycled Tinder UI’s have been sending pretty dire warning signals that they are seriously struggling to carry on into the “3rd wave of dating apps”. Alike companies and their sudden increase in monetization features really represents this issue for our predecessors. The most recent signal being “Tinder Gold” which allows users to view who liked you or Bumble’s “bumble boost” feature. We do owe Tinder and other apps a great deal of respect for expanding dating app DAU across the globe. But these apps with very impressive amounts of daily active users are absolutely desperate to make money and are in fact burning through huge budgets barely turning a single profit. As these companies continue to scramble for ways to reverse the trend of “free” casual dating apps that users expect no restrictions on and fail to deliver, its naturally evident that the hot or not, casual and free style of dating apps will be expiring sooner than never. Data shows us that millennials are actually very happy and capable of paying for dating apps that get first dates. This is why we have been seeing so many people hopping on next-gen dating apps like Dine who has amassed an incredible user base in just one year since launch. The trend of Niche dating apps will die Every day it seems as if a new dating app emerges taking an attempt to solve specific issues that cater to certain groups of users. Here are just a few examples from the 100’s of niche dating apps/sites available now: Align — A dating app that matches you with people based on your zodiac signs. Famers Only — A dating site that connects single farmers looking for a shot at love. The League — A dating app that prides itself on an app with the most affluent and intelligent dating pool. To even get accepted into the community there is an initial screening process and requirements to meet. The death of these niche apps: Niche apps may play well into the hands of a few when it comes to matching on aspects like race, religion, gender. However, the trend of dating apps that match you based on a specific similarity (sometimes obscure), these concepts come up short. Most are nothing short than a joke or a clever icebreaker to open up conversation with someone new. A great recent example is a dating app that launched a few months back called “Hater”, another recycled Tinder-like UI that matches you based on the “things you hate”. A user and their match receives a compatibility score that’s measured through a feature that allows users to swipe love, like, dislike, and hate. Another example could be “Tastebuds”, a dating app that allows you the meet new people through similar tastes in tunes. All of these niche apps sound cute and unique on paper but in reality they fail miserably in focusing on what really matters to users, getting out on actual dates. Just because Jane hates peanut butter and Joe like peanut butter doesn’t mean they couldn’t be the happiest couple in the world. Just because Billy loves heavy metal and Nancy can’t stand it doesn’t mean they couldn’t be perfect for each other. From a business side, they will never be able to scale large amounts of DAU compared to what most of us are used to when you cut off large percentages of groups to join the user pool (loses a lot of monetization opportunities too). Unless these apps get more innovative and keep their eyes on the prize, many will most likely start drying up one by one, a process that has already begun; natural selection if you will. A.I. Yup, robots and artificial intelligence will certainly become a strong asset to dating app developers and users alike. AI research and development continues to make huge steps every day and some dating app developers are already brainstorming and testing out how they can incorporate AI into dating apps. What’s the need for integrating AI in dating apps one may ask? Simply put, AI has the power to increase a user’s experience. For example, Dine, the dating app that guarantees first dates, uses a “smart preference feature” which on a weekly basis requests all users to either “like” or “dislike” a series of 100 random photos across a diverse pool of users with the main purpose being to calibrate specific tastes of a user. If dating apps can learn naturally through daily usage of what a user’s preferences are, apps like Dine can can be more successful by introducing users to higher quality matches. This feature is supported by a listing logic algorithm, but how accurate is this listing logic actually? Listing logic does its best to take notes about specific details and patterns/trends of a user’s activity (what they are attracted to in this case). In reality, listing logic is not perfect and only reaps visible results when used consistently over a long period of time. As we dive into the future, incorporating AI into dating apps can seriously hunker down on the little details that cater to our own personal attractions. Only like men tall men? Sure. Only like blonde haired women? No problem! Blue eyed people? OK! AI has the power to efficiently and effectively learn about a user’s preferences in great detail which in return can significantly increase the user’s experience on dating apps without having to exclude too many potential matches with blanket type algorithms and filters. Cool right? Virtual Reality That’s right. Companies like oculus are making huge strives with VR tech and it will only go up from here. We’re already seeing the company playing around with Dating Lessons courses which covers topics like “how to approach women”. These are just the beginning steps of VR and online dating but imagine, a world where there are actual retail stores set up that allow people to sit in booths and meet and chat with available singles both local and from all over the world at any time of the day. Pretty crazy right?…Some may even argue sad, but this is certainly towards where we’re heading as online safety and privacy become more and more crucial to us. Many may even say that this model is a safe and effective way to screen out people in a fast paced online dating world where options are endless but some potentially dangerous. Conclusion Although it is hard to predict the future, one thing that is for certain is that the dating app industry is a 4-billion dollar industry that has been growing at a steady pace of 5% year-over-year since 2013. More and more people are using dating apps and meeting their partners which has been essentially erasing the dating app stigma completely. In the USA alone, 30% of people ages 18+ are existing users or have used a dating app/online service at least once in their lifetime. Leave a comment about what you think the next trend will be! Check out: I used Tinder, Happn, Bumble and Dine for 2 weeks. Results were interesting.

Where do we go from here? Tinder and the future of dating apps

24

where-do-we-go-from-here-tinder-and-the-future-of-dating-apps-124ada7edc60

2018-05-07

2018-05-07 21:52:59

https://medium.com/s/story/where-do-we-go-from-here-tinder-and-the-future-of-dating-apps-124ada7edc60

false

1,421

null

Dating

dating

Dating

24,079

Michael Tudda

Marketing Director/Foodie/Tokyo/NYC

1bb54397f794

DineWithMike

129

299

20,181,104

null

0

iris = sklearn.datasets.load_iris() X = iris.data[:, :2] y = (iris.target != 0) * 1 def sigmoid(z): return 1 / (1 + np.exp(-z)) z = np.dot(X, theta) h = sigmoid(z) def loss(h, y): return (-y * np.log(h) - (1 - y) * np.log(1 - h)).mean() gradient = np.dot(X.T, (h - y)) / y.shape[0] lr = 0.01 theta -= lr * gradient def predict_probs(X, theta): return sigmoid(np.dot(X, theta)) def predict(X, theta, threshold=0.5): return predict_probs(X, theta) >= threshold class LogisticRegression: def __init__(self, lr=0.01, num_iter=100000, fit_intercept=True, verbose=False): self.lr = lr self.num_iter = num_iter self.fit_intercept = fit_intercept def __add_intercept(self, X): intercept = np.ones((X.shape[0], 1)) return np.concatenate((intercept, X), axis=1) def __sigmoid(self, z): return 1 / (1 + np.exp(-z)) def __loss(self, h, y): return (-y * np.log(h) - (1 - y) * np.log(1 - h)).mean() def fit(self, X, y): if self.fit_intercept: X = self.__add_intercept(X) # weights initialization self.theta = np.zeros(X.shape[1]) for i in range(self.num_iter): z = np.dot(X, self.theta) h = self.__sigmoid(z) gradient = np.dot(X.T, (h - y)) / y.size self.theta -= self.lr * gradient if(self.verbose == True and i % 10000 == 0): z = np.dot(X, self.theta) h = self.__sigmoid(z) print(f'loss: {self.__loss(h, y)} \t') def predict_prob(self, X): if self.fit_intercept: X = self.__add_intercept(X) return self.__sigmoid(np.dot(X, self.theta)) def predict(self, X, threshold): return self.predict_prob(X) >= threshold model = LogisticRegression(lr=0.1, num_iter=300000) %time model.fit(X, y) CPU times: user 13.8 s, sys: 84 ms, total: 13.9 s Wall time: 13.8 s preds = model.predict(X) # accuracy (preds == y).mean() 1.0 model.theta array([-25.96818124, 12.56179068, -13.44549335]) model = LogisticRegression(C=1e20) %time model.fit(X, y) CPU times: user 0 ns, sys: 0 ns, total: 0 ns Wall time: 854 µs preds = model.predict(X) # accuracy (preds == y).mean() 1.0 model.intercept_, model.coef_ (array([-80.62725491]), array([[ 31.61988897, -28.31500665]]))

18

null

2018-02-21

2018-02-21 15:21:57

2018-02-23

2018-02-23 00:04:19

6

false

en

2018-02-23

2018-02-23 18:58:43

2

124c5636b8ac

3.878302

90

7

0

While Python’s scikit-learn library provides the easy-to-use and efficient LogisticRegression class, the objective of this post is to…

5

Logistic Regression from scratch in Python While Python’s scikit-learn library provides the easy-to-use and efficient LogisticRegression class, the objective of this post is to create an own implementation using NumPy. Implementing basic models is a great idea to improve your comprehension about how they work. Data set We will use the well known Iris data set. It contains 3 classes of 50 instances each, where each class refers to a type of iris plant. To simplify things, we take just the first two feature columns. Also, the two non-linearly separable classes are labeled with the same category, ending up with a binary classification problem. Fig. 1 — Training data Algorithm Given a set of inputs X, we want to assign them to one of two possible categories (0 or 1). Logistic regression models the probability that each input belongs to a particular category. Hypothesis A function takes inputs and returns outputs. To generate probabilities, logistic regression uses a function that gives outputs between 0 and 1 for all values of X. There are many functions that meet this description, but the used in this case is the logistic function. From here we will refer to it as sigmoid. Fig. 2 — Logistic function Fig. 3— Hypothesis Loss function Functions have parameters/weights (represented by theta in our notation) and we want to find the best values for them. To start we pick random values and we need a way to measure how well the algorithm performs using those random weights. That measure is computed using the loss function, defined as: Fig. 4— Loss function Gradient descent Our goal is to minimize the loss function and the way we have to achive it is by increasing/decreasing the weights, i.e. fitting them. The question is, how do we know what parameters should be biggers and what parameters should be smallers? The answer is given by the derivative of the loss function with respect to each weight. It tells us how loss would change if we modified the parameters. Fig. 4 — Partial derivative Then we update the weights by substracting to them the derivative times the learning rate. We should repeat this steps several times until we reach the optimal solution. Predictions By calling the sigmoid function we get the probability that some input x belongs to class 1. Let’s take all probabilities ≥ 0.5 = class 1 and all probabilities < 0 = class 0. This threshold should be defined depending on the business problem we were working. Putting it all together Evaluation Fig. 5 — Decision boundary Picking a learning rate = 0.1 and number of iterations = 300000 the algorithm classified all instances successfully. 13.8 seconds were needed. These are the resulting weights: LogisticRegression from sklearn: If we trained our implementation with smaller learning rate and more iterations we would find approximately equal weights. But the more remarkably difference is about training time, sklearn is order of magnitude faster. Anyway, is not the intention to put this code on production, this is just a toy exercice with teaching objectives. Further steps could be the addition of l2 regularization and multiclass classification. Code available here.

Logistic Regression from scratch in Python

486

logistic-regression-from-scratch-in-python-124c5636b8ac

2018-06-16

2018-06-16 11:41:23

https://medium.com/s/story/logistic-regression-from-scratch-in-python-124c5636b8ac

false

776

null

Machine Learning

machine-learning

Machine Learning

51,320

Martín Pellarolo

Data Scientist and Software Engineer | Passionate about AI — Machine Learning — Deep Learning.

2d1b498ccdb5

martinpella

101

6

20,181,104

null

0

null

0

32881626c9c9

2018-08-09

2018-08-09 16:09:46

2018-08-10

2018-08-10 16:51:08

1

false

en

2018-08-21

2018-08-21 01:32:11

1

124d1de9307d

3.977358

5

0

I hope this is a worth-while innovation on Geoff Hinton’s Capsule networks. Rather than using strict ‘orientation’ vectors, which Hinton…

5

Learning Capsules I hope this is a worth-while innovation on Geoff Hinton’s Capsule networks. Rather than using strict ‘orientation’ vectors, which Hinton relied upon, I suggest using ‘context’ vectors which are learned. Additionally, when contexts do not align at a higher layer of the network, I offer a method for resolving dissonance. Key to these developments is the use of the entire distribution of possible interpretations, and a memory of when unusual interpretations are used. Interpretations and Context Suppose I give a neural network an image of an orange sphere. The network may interpret this blob of color as a pumpkin, or a basket ball, or a chew toy, or… There are many reasonable interpretations. While most neural networks throw away this distribution, returning only the most heavily weighted response, (“pumpkin!”) my method retains the distribution of interpretations — it remembers ‘pumpkin’, ‘basket ball’, ‘chew toy’, etc. This is necessary because, in our new network, dissonance at a higher layer will trace backwards to re-try the feed-forward step using secondary interpretations. If ‘pumpkin’ doesn’t make sense, given the picture’s context, then the network sees if ‘basket ball’ makes sense, instead. The interpretation module takes the input, and converts it into a list of all appropriate interpretations. When the top interpretation doesn’t make sense, the next interpretation is chosen. In addition to the power of selecting a more reasonable interpretation, the suggested network also has freedom to choose how it represents the context of the interpreted image. Similar to Hinton’s ‘orientation’ vector, this context vector declares how the ‘archetypal’ interpretation is modified, to produce the observed inputs. Hinton’s ‘orientation’ vector declared where numbers were positioned on a page, as well as any rotation applied to the number, allowing the network to regenerate the input faithfully from its interpretation (the number chosen) and its orientation. Our context vectors have the freedom to mean more than just ‘position and orientation’ — they are learned alongside the module that generates a distribution of interpretations. Together, these interpretations and their contexts must be sufficient to regenerate the observed input. The context vector might describe the texture and color of the background, across large swatches, as well as the position and orientation of the object. The network would see ‘pumpkin on dead grass’ or ‘chew toy on dog fur’, for example. There are further complications, though — the process of selecting whether two context vectors are similar enough to be combined, and the process of combining them into a higher-layer concept. Hinton’s ‘orientation’ vectors were static, and they could be compared to each other using a hard-coded similarity. Because our context vectors will use learned representations, some dimensions of those representations may be more important than others, in specific settings. Our context-comparator must be another neural network module, which receives a group of contexts and outputs whether or not those contexts are in sufficient agreement for them to be grouped into a higher concept. Like Hinton’s ‘orientation’ vectors, a lack of agreement between context vectors implies that the selected interpretations are poor. That’s when the secondary interpretations are employed, instead. Should context vectors agree, then the interpretations found among various modules are combined into a higher-layer concept. Swatches of shaggy brown and dimpled orange combine to form ‘dead lawn’ and ‘pumpkin’. Together, ‘pumpkin’ and ‘dead lawn’ build to the concept of ‘Halloween’, allowing a neural network trained for describing scenes to decide upon the description: ‘a group of trick-or-treaters walk between houses.’ The higher concepts of Halloween and trick-or-treater are interpretations derived from the lower-layer interpretation as ‘pumpkin’. These higher-layer concepts only occur if the ‘pumpkin’ interpretation matches the context from other areas of the image. The method for comparing these contexts is not straight-forward like Hinton’s ‘orientation’ vectors — they must be learned. When the distribution of interpretations, the context vector, and the context-comparator have all been trained, the network should be able to regenerate the original input, using its memory of which interpretation was selected for each object and the scene as a whole. Back-propagating through Choices To train this suggested network, we must compare its output to the correct answer, and back-propagate through the chain of modules which generated the output. That means, if a secondary interpretation is used, we back-propagate through the network when it is set to that secondary interpretation. The memory used to regenerate the input is the same memory necessary for properly assigning the gradient through back-propagation. Those memories keep which interpretation was chosen; we are back-propagating through a choice. That choice should not be replaced by a distribution across possible choices. The initial interpretations in that distribution were abandoned because they generated conflicting contexts at higher layers, and later interpretations may or may not be relevant. Only the interpretations that were actually chosen should be subject to back-propagation. Yet, this back-propagation alters the context-generating module, which may alter many unrelated contexts and interpretations. I do not pretend that such a network would be easy to train. Rather, it might learn quite slowly, because any error in the distribution of interpretations, or in the context-generator, or in the context-comparator could create an observed error in output. Training may be even more difficult, because choosing the interpretation which minimizes dissonant contexts results in a loss-landscape with discontinuities. A slight change to the distribution of interpretations might introduce a new interpretation, leading to a cascade of new interpretations at higher layers. Similarly, altering the context-generator or context-comparator might yield disagreement which leads to selecting a different interpretation. Yet, I see this structure, which uses learned contexts and learned comparisons, to be superior to Hinton’s — it is a generalization of the process found in his capsules, which can take a broader set of inputs and operate upon them in a diverse set of ways. Not just for imagining the displacement and rotation of numbers!

Learning Capsules

51

learning-capsules-124d1de9307d

2018-08-21

2018-08-21 01:32:12

https://medium.com/s/story/learning-capsules-124d1de9307d

false

1,001

Data Driven Investor (DDI) brings you various news and op-ed pieces in the areas of technologies, finance, and society. We are dedicated to relentlessly covering tech topics, their anomalies and controversies, and reviewing all things fascinating and worth knowing.

null

datadriveninvestor

null

Data Driven Investor

info@datadriveninvestor.com

datadriveninvestor

CRYPTOCURRENCY,ARTIFICIAL INTELLIGENCE,BLOCKCHAIN,FINANCE AND BANKING,TECHNOLOGY

dd_invest

Machine Learning

machine-learning

Machine Learning

51,320

Anthony Repetto

Easily distracted mathematician

6374f82a1f5c

oaklandthinktank

673

97

20,181,104

null

0

null

0

null

2018-02-19

2018-02-19 13:26:15

2018-02-19

2018-02-19 15:30:12

0

false

en

2018-02-19

2018-02-19 15:30:12

1

124d1e4bbe16

0.060377

0

null

5

Machine Learning Better Protecting Minority Investors doing business ml doing business ml | Piktochart Visual Editorcreate.piktochart.com

Machine Learning Better Protecting Minority Investors

0

machine-learning-better-protecting-minority-investors-124d1e4bbe16

2018-02-19

2018-02-19 15:30:13

https://medium.com/s/story/machine-learning-better-protecting-minority-investors-124d1e4bbe16

false

16

null

Business

business

Business

153,000

WELTARE Strategies

WELTARE Strategies is a #startup studio raising #seed $ for #sustainability | #intrapreneurship as culture, #integrity as value, @neohack22 as Managing Partner

9fad63202573

WELTAREStrategies

196

209

20,181,104

null

0

null

0

null

2018-02-08

2018-02-08 13:48:36

2018-02-28

2018-02-28 08:01:01

2

false

en

2018-02-28

2018-02-28 08:01:01

0

124db08de1bb

2.700314

2

0

1

Fraud is one of the ancient things in human history. As there is always people who are fraudulent, there is also people who defrauded. The…

4

Fraud Detection with ML Fraud is one of the ancient things in human history. As there is always people who are fraudulent, there is also people who defrauded. The money e.g. credit card information is one of the well-known targets for being targeted by fraudulent activities. With the development of e-marketing sector, the count of fraudulent activities is rising day by day. Users credit cards information stored in some companies’ databases, such company types as banks, online shopping companies or online service providers. We witness a growing presence of frauds on online transactions with the widespread use of internet day by day. As a consequence of this, the need of automatic systems which able to detect and fight fraudster has emerged. Fraud detection is notably a challenging problem because; Fraud strategies change in time, as well as customers’ spending habits evolve. Few examples of frauds available, so it is hard to create a model of fraudulent behavior. Not all frauds are reported or reported with a large delay. Few transactions can be timely investigated. With the large number of transactions we witness everyday; We can not ask the human analyst to check every transaction one by one. We wish to automatise to detection fraudulent transaction. We want the accurate prediction, i.e. minimize missed frauds and false alarms. It can be overcome this bad situation with systems which developed with machine learning techniques. Systems can learn complex fraudulent pattern by examining the data in large volumes. And these systems can also create an optimal model for fraudulent activities which has complex shapes. Thus, successful predict can be done for the new type of fraud. And the system can adapt itself to timely changing distribution (fraud evolution). However, systems need enough samples to achieve successfully learning. Basically, user profile created for every user in the detection systems. This profile must be updated timely. When the system has trained with enough samples, systems has detailed information about users spending habits for monthly, weekly or daily. For example, suppose that while a student can spend $100 for a week, a businessman can spend $1000 for a week. While a fraudulent activity with $400 spends at once has high fraud probability for a student, similarly, it has very low fraud probability for a businessman. Of course, special days such as new year, birthday or weekends must be considered when creating an algorithm for fraud detection, because students can also spend too much in these days. Generally, fraudulent does not know victim’s spending habits, because of that fraudulent activity has inappropriate matching to user profile presumably. But if fraudulent activity fits user profile, it may be hard to detect. In machine learning literature, fraud detection systems can be built with supervised, unsupervised or mixed approaches. Every type of approach has a little difference according to working logic. Approaches and their working logics have given Figure below. In supervised learning, using labeled historical fraud data to create a user profile in the training phase. This type of approaches are similar to Signature Based IDS/IPS, so this type of approaches can detect fraudulent activities if they are well known, but a new type of fraudulent activities can not be detected by these systems. Systems which is trained with unsupervised algorithms can detect unknown fraudulent activities. This type of approaches are similar to Anomaly Based IDS/IPS, so although this type approaches can detect unknown fraudulent activities, in some cases they may not detect well known fraudulent activities. To achieve the disadvantages of both techniques, mixed approaches are developed. In this type of approaches, supervised and unsupervised algorithms work together, so both well-known and unknown fraudulent activities can be detected efficiently.

Fraud Detection with ML

2

fraud-detection-with-ml-124db08de1bb

2018-02-28

2018-02-28 15:42:06

https://medium.com/s/story/fraud-detection-with-ml-124db08de1bb

false

614

null

Machine Learning

machine-learning

Machine Learning

51,320

Ebubekir Büber

null

ff57f208d81d

EbubekirBbr

154

54

20,181,104

null

0

null

0

null

2018-03-13

2018-03-13 16:57:54

2018-03-13

2018-03-13 17:07:40

1

false

en

2018-03-13

2018-03-13 17:07:40

6

124ed09d55f3

2.203774

0

Cryptocurrency investment is not something weird nowadays. People tend to like it not only for the benefits to be gained but also for the…

5

The Best Place for Cryptocurrency Trading and Investment Cryptocurrency investment is not something weird nowadays. People tend to like it not only for the benefits to be gained but also for the possibility that this will be long lasting. In the future, it is forecasted that digital currency will still be developed. The users will be more and more as well. So, why don’t you try just from now? However, the way how to begin is not something easy indeed. It can also be said as really puzzling. One of them is regarding what company or provider to choose. Many options are available out there but it seems that Signals is one of the most recommended. Why? Here is the reason. Machine Intelligence There is no 100% guarantee of being profitable in the world of investment. But it is a duty for the investors to think more deeply so that what they have invested can generate good results. Unfortunately, it is somehow difficult to do for the exception if you are experienced enough. Well, Signals really know about this matter. Therefore, to help you in predict and do anything else related to this investment, the machine intelligence is provided. So, what is it actually? In this cryptotrading industry, even each trader is equipped by the tools so-called the computational power and data science. Therefore, it can just make the trading is done in much smarter ways. Besides, the process tends to be faster as well with it. There is indeed no time to learn more about the algorithm and so forth. So, if there is a helpful tool to deals with those matters, why don’t you have to make it more difficult? Easy Access For common people, cryptotrading and cryptocurrency may be something that is out of their reach. It is not about they don’t have enough information and the likes. It is only related to its difficult vibe. As a result, many people seem afraid to start it even if they really want to. In Signals, anything becomes much easier. It is very easy to access and learn. First of all, you need to read the Whitepapers or the terms and conditions about what to do and what must not do. Interestingly, it is available in many languages aside from English like Japanese, Chinese, Spanish, Russian, Portuguese, and Korean. So, everybody can just understand it more easily. The first step to join this Cryptotrading industry is also simpler through Signals. You don’t need to be very experienced for this. Just make yourself sure that you want to study more about this and the success is just on your hands. Of course, with machine intelligence mentioned above, it is much easier. Share the Strategies It is much better if all the members are able to share the strategies and experiences here. The success can just be reached together anyway. Interestingly, you can also monetize it. it is by placing your invention of new trading model in Signals Marketplace. So, are you interested in it? be sure to check out https://signals.network for further information. ° Website ° Facebook ° Twitter ° Whitepaper ° Announcement on Bitcointalk Author: CryptoSmile ETH Address: 0x0123a524a2d5A69d9E934750951a519dcB552E59

The Best Place for Cryptocurrency Trading and Investment

0

the-best-place-for-cryptocurrency-trading-and-investment-124ed09d55f3

2018-03-13

2018-03-13 17:07:42

https://medium.com/s/story/the-best-place-for-cryptocurrency-trading-and-investment-124ed09d55f3

false

531

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Crypto Smile

null

da63717a71de

cryptosmile

199

238

20,181,104

null

0

null

0

db26a4b955f9

2018-05-02

2018-05-02 11:46:14

2018-05-02

2018-05-02 11:53:12

2

false

en

2018-05-02

2018-05-02 11:53:12

1

124f6c3c2013

7.417296

2

0

Data Science: Where to start?

5

How to leverage Data Science to increase growth? Data Science: Where to start? The dashboard experts and excel ninjas within your organization, with skills that worked so well a decade ago, can only take you so far in this age. Most have come to accept this. We’re all data rich today, so why are we finding it so difficult to drive our growth backed by data? Consider crude oil. Without refining it, this energy powerhouse is nothing more than brown liquid. In today’s world, where data is increasingly considered the king of resources, many organizations are collecting vast amounts of data in the name of giving it importance, but gleaning very little information out of it. While an increase in data volume gives you more to analyze, most are struggling with how to make use of it. It’s as if organizations are sitting on enormous oil reserves, but lack the infrastructure to turn it into anything truly useful. They are data-rich but insight-poor. The key to unlocking the entire potential of data lies in the data itself and the ones employed to make use of it. Below are a few pointers to help you understand the gaps you need to fill before leveraging the power of data to drive growth. Data science serves two distinct goals, improving the products your customers use and improving the decisions your business makes. What is your priority? To improve product performance, we employ a cyclical process where products collect usage data that becomes the fodder for algorithms which in turn offer users a better experience. For this to be successful, data scientists will have to closely collaborate with engineers. When it comes to driving business decisions, it’s important to understand that every decision doesn’t require the top talent of decision science. Certain decisions are too small to justify the investment. Other decisions may be important, but you might lack the data to meaningfully analyze them. Here, you’ll have to rely on experimentation and intuition. It’s rare for data scientists to excel at both, even though decision science and data products call for some of the same skills. Decision science depends on systems thinking, business and product sense, and strong communication skills. Data products require production-level engineering skills along with machine learning knowledge. If you’re starting out with a small team, you may have to hunt for the rare ones who can do both. But you’ll see the advantages of specialization as you scale. An engineer can create a Minimum Viable Product (MVP) with a small amount of design and product guidance. Data science will need data that only comes with scale and measurement. This will require a commitment across the organization to identify what data each product needs to collect and along with the establishment of processes and infrastructure for collecting and maintaining that data. To do this successfully, it requires collaboration among engineers, product managers and data scientists, which demands executive commitment. Do not rush into hiring because you think you’re missing out and believe you have the necessary infrastructure ready for the incoming data scientist. Too many companies feel a sense of urgency when it comes to building a data science team. Companies with modest ambition are eager to hire costly folks who will derive insight from all that data. Building a team too early can prove to be an expensive distraction and will demotivate your talent and might have abiding negative cultural ramifications. Even if you see yourself becoming increasingly data-reliant as you grow, don’t succumb to making hasty hires and hastier fires. It takes time to find the right people for your organization and time for the hires to get to know your business and your data. All this needs to happen before you can apply data science to drive decision making within your business. It’s time to consider investing in data science if you’re done validating your MVP. A successful product launch must’ve generated enough data to learn from. Bring people on board who can keep up with that data stream by extracting value and insight from it. Now before hiring, ask yourself whether data science is a core competency or whether you can afford to outsource it? Building a competent data science team will prove to be hard and expensive, even for giants in the business. If you can outsource your data science needs and get away with it, go for it. You can make use of off-the-shelf solutions for your domain that makes use of APIs to ingest data, automate actions, build models and report on key analytics. Don’t expect to find the perfect fit, but often, it’s worth compromising here to accelerate your business and keep your core team focused on areas where it can add the most value. However, it’s a risk to outsource when data science is at the heart of your business. Also, off-the-shelf solutions are usually inflexible. If you plan on approaching a problem in a unique way like using the results in novel ways or collecting new kinds of data, an off-the-shelf solution most probably won’t be flexible enough to adapt to it. Where you introduce data science in your organizational structure matters a lot. You can choose to build a data science team that functions as an independent team along with engineering. Here, the lead data scientist would report to the head of engineering or to the CEO, directly. In this type of structure, the data science team is positioned such a way that they can work on any issue they deem most critical. This structure also demonstrates that you see data as a first-class asset, which will help you attract world-class talent. But here, you risk marginalization. As companies grow in number, each team often prefers to be self-sufficient. Generally, product teams simply don’t want to depend on resources they can’t control, even when they can take advantage of collaboration with data scientists. If this happens, your data team will soon become ineffective and marginalized. You can also introduce data science in a way where you recruit a team that brings in talented data people and farms them out to the rest of the company as needed. Here, your head of data science basically functions as a coach and hiring manager. In this model, to ensure utility, your data team lets go of autonomy. In the best case scenario, data scientists end up joining product teams that need their service the most and get to work. The downside is, not all data scientists are happy giving up autonomy. So, you can choose to adopt the third model where product teams hire and manage their own data scientists. This will help with organizational alignment. Data scientists, product managers, designers, and software engineers end up working together on shared product goals. With this model, you can instill a collective sense of ownership. But here, rather than having a centralized data science team, you choose to dilute your product team. And since it can be hard to move people around based on their skills and interests from a team to another, you also sacrifice flexibility. Each path has its upside and downside. Remain willing to adapt as your needs change, and it doesn’t hurt to mix things up. Often, the best approach is never textbook one, so adopt a hybrid if it works for you. Sometimes we get so caught up in managing and collecting data that we lose sight of why we are expending so much time and effort in doing so. Data is only as important as the action it drives. Data should drive your organization’s key performance indicators (KPIs) and inform product changes. Too many businesses today claim to be data-driven. These companies collect an insane amount of data, invest heavily in data engineering, and reference data heavy dashboards frequently. Yet, they fall short. You have to begin by constructing the backbone and the credibility crucial to making decisions based on data even when they lead to significant power shifts in your business or run counter to mainstream wisdom. Only when you commit to a culture of data-driven decision making can data science leave a lasting impact on your company. A top-down commitment is crucial in instilling a data-driven culture. From the CEO down, your company has to commit to making decisions based on data rather than on the highest paid person’s opinion. It’s never too early to instill a data-driven culture. An advantage of introducing data science into an organization earlier on is that doing so ensures everybody begins to view data as a first-class asset from the word go. And it’s essential that every team in your company respects data because a data scientist in a vacuum can achieve nothing. It takes a village to drive growth through data. Unless data scientists, product managers, engineers, and designers collaborate closely, it’s impossible to create amazing products. Unless leaders and executives value insights gleaned from data, a data scientist’s recommendations would never affect change. Build a diverse team coming from different backgrounds, skill-sets, and world-views over time, this way the impact that a data science team has will be far higher as diverse teams tend to think as holistically as possible about their domain, and will ideally encourage creativity and innovation in time to come. Most young data scientists walk into the field with the belief that they will get to solve complex problems with fancy machine learning algorithms that will eventually transform the business. They see the stream as a budding field where they stand a chance to feel like their role leaves an impact. However, this is often not the case. The reason we see attrition in this job role is because expectations rarely match reality. Too many of us hire data scientists before setting the right infrastructure in place. On top of that most of us choose to hire juniors before hiring experienced data practitioners and you’ve got a recipe for a disillusioned, unhappy relationship for both sides. The data scientist most probably joined the team to help script smart machine learning algorithms to glean insight but isn’t able to get to do that yet since now he or she has to first set up the data infrastructure and built analytic reports. The company, on the other hand, was only looking for a fancier chart to present at their weekly meetings every Monday. The company then gets frustrated because they fail to see value being driven fast enough and all this ends up with the data scientist unhappy in their role. Don’t be that company. We have seen enough of the good and the ugly while advising and leading teams at various companies at different stages of maturity from different industries. We’ve seen the challenges of not just driving growth by relying on data but also of hiring top data scientists and retaining them in today’s market. If you’re looking for a partner on your data-analytics journey, we’d be happy to jump on board. Originally published on Product Insights Blog from CognitiveClouds: Top Ruby on Rails Development Company

How to leverage Data Science to increase growth?

6

how-to-leverage-data-science-to-increase-growth-124f6c3c2013

2018-05-25

2018-05-25 08:07:37

https://medium.com/s/story/how-to-leverage-data-science-to-increase-growth-124f6c3c2013

false

1,864

Tips, advice and insights from our digital product strategy, design and development experts.

null

CognitiveCloudsLLC

null

CognitiveClouds

null

cognitiveclouds

WEB DEVELOPMENT,MOBILE APP DEVELOPMENT,SOFTWARE DEVELOPMENT,IOS APP DEVELOPMENT,ANDROID APP DEVELOPMENT

cognitiveclouds

Data Science

data-science

Data Science

33,617

Amit Ashwini

VP of Marketing at Zibtek. I help top startups and companies build remarkable web, mobile and tablet products.

b9437eb132f6

amitashwini

803

247

20,181,104

null

0

null

0

d777623c68cf

2017-10-14

2017-10-14 11:31:10

2017-10-14

2017-10-14 11:36:00

4

false

en

2018-01-24

2018-01-24 17:08:03

5

1250a437cdb4

4.824528

522

26

1

Do you know what’s more dangerous than artificial intelligence? Natural stupidity. In this article, I will explore natural stupidity in…

3

Natural Stupidity is more Dangerous than Artificial Intelligence Support my writing by treating yourself with a t-shirt: https://teespring.com/shop/natural-stupidity-is-danger Do you know what’s more dangerous than artificial intelligence? Natural stupidity. In this article, I will explore natural stupidity in more detail and show how our current technology (driven by narrow artificial intelligence) is making us collectively dumber. We’ve all had this experience of using a GPS to guide us around an unfamiliar place only to realize later that we have no recollection or ability to get to that place again without the aid of a GPS. Not only is our directional instinct diminished because of lack of use, but so is our own memories. We’ve all experienced losing our ability to recall due to our over use of Google. We now recall more as to how we can search for something rather than the details of that something. The framework that I often use to explore intuition is the Cognitive Bias Codex found at Wikipedia. It’s a massive list of biases, however to get an overview of it, there are four high level categories that are the the drivers of theses biases. These are “Too Much Information”, “Not Enough Meaning”, “Need to Act Fast” and “What Should we Remember?”. Source: https://en.wikipedia.org/wiki/List_of_cognitive_biases Our world requires more automation to run efficiently and sustainably. The products and services that will be in demand are the products that compensate for our inadequacies. The clear downside of this is that with every assist, the less we exercise our already weak facilities. The only people maintaining their smarts are the few people willing to constantly exercise their smarts. Meanwhile, we have a population that is becoming more out of shape and lazy with their own mental faculties. We imagine ourselves to being smarter because we can multi-task more. Yet, our brains have not evolved to do multi-tasking well. In fact, recent research have shown that pigeons have greater multi-tasking capabilities than humans. It is just ironic that we’ve taken pride in our new found multi-tasking skills only to discover that we are dumber at it than pigeons! However, there is a far worse problem than automation making us dumber. The bigger problem is that other humans are aware that it can make us dumber and they are opportunistically exploiting our natural stupidity to influence our behavior. Over the decades, the industry of advertising has spend trillions of dollars inventing new ways to “motivate” us to do new things without us being aware of its influence. The techniques to do this neatly falls under the exploitation of our cognitive biases. After all, if we were indeed all perfectly logical, then we’ll likely spending our money in the most efficient way possible and very few companies will like us to do that. If we reduce our spending, our economies would stall and there would be an economic depression! (BTW, something is really wrong when we must accelerate our consumption so as to avoid economic stagnation) So, “Natural Stupidity” is basically our lack of meaning, lack of memory, inability to think fast and inability to process too much information. The current systems that we have in place provide products and services to substitute these inabilities. It is the natural tendency to seek out the method of least action. That is, the method that requires lest effort or the laziest thing that we can do. Let’s explore each of the four in greater detail. Humans from the beginning of the their life are driven to seek meaning. The simplest explanations to this are going to be the most natural appealing ones. Civilization will naturally create religion to not only create a necessary shared understanding of acceptable behavior but one that is driven by our need for meaning. The written word (i.e. books) and its more advanced form, the world wide web are devices that address our limited memories. Memories require not only storage but also the capability of recall. Throughout history, religion and law has been transcribed in scrolls, books and now in automation (see: blockchain). Money is a form of memory, that is, once possession of it is a measure of one’s ability to acquire goods and services. Mankind created computers to automate the math that we invented. Computers not only store memories but are able to perform laborious and error-free computations. We find it an inconvenience to use cash in that we have to calculate in our minds the amount of change so as to guard against error or outright fraud. We have time keeping devices so that we don’t need to look out into the heavens to determine the time of day. We have GPS devices to help us avoid reading a map and calculating a path to our destination. Finally, we have the problem of information overload. Our knowledge driven economies have accelerated our consumption of information. However, our brains have not magically evolved to process this fire-hose of information. The device that we use to process more information are services that curate information and exhaust it out in more easily digestible forms. Today, social networks such as Twitter and Facebook have become our primary tools for curating and receiving new information about the world. It is dumfounding that the leaders of these two companies believe it is not in their charter to ‘police’ the contents that they help propagate. With great power comes great responsibility, unless it I guess if it affects the bottom line! We collectively become dumber when we relinquish responsibility and accountability to the automation (or A.I.) that furnishes us with cognitive assistance. When we avoid questioning the positions of our religious leaders and ignore obviously repugnant behavior in defense of our own beliefs. We avoid verifying our history and cling to untrue historical information to justify our beliefs. This is the case for Neo-Nazis and Confederates who would like to imagine a more benevolent and just past. We ignore common sense by following algorithms in enforcement of procedures. Like the United Airlines where a 70 year old doctor was assaulted and removed from an airline just because the crew blindly followed protocol instead of their own common sense. Finally, we don’t hold accountable organizations that employ information overload in the form of massive disinformation to mold public opinion. We shouldn’t be worried about Artificial Intelligence taking over the world. The more immediate, clear and present danger is that Natural Stupidity has taken over this world and we are seeing daily occurrences of this in our public discourse. More on how intuition, human behavior and artificial intelligence are tightly related: . Explore Deep Learning: Artificial Intuition: The Unexpected Deep Learning Revolution Exploit Deep Learning: The Deep Learning AI Playbook

Natural Stupidity is more Dangerous than Artificial Intelligence

2,919

natural-stupidity-is-more-dangerous-than-artificial-intelligence-1250a437cdb4

2018-06-01

2018-06-01 16:52:22

https://medium.com/s/story/natural-stupidity-is-more-dangerous-than-artificial-intelligence-1250a437cdb4

false

1,093

Deep Learning Patterns, Methodology and Strategy

null

deeplearningpatterns

null

Intuition Machine

info@intuitionmachine.com

intuitionmachine

DEEP LEARNING,ARTIFICIAL INTELLIGENCE,MACHINE LEARNING,DESIGN PATTERNS

IntuitMachine

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Carlos E. Perez

Author of Artificial Intuition and the Deep Learning Playbook — Intuition Machine Inc.

1928cbd0e69c

IntuitMachine

20,169

750

20,181,104

null

0

null

0

98e37200303a

2017-12-10

2017-12-10 14:51:53

2017-12-11

2017-12-11 04:42:38

3

false

en

2017-12-11

2017-12-11 04:42:38

1

1252958e3bf0

1.663208

0

This week we decided our datasets which we can use in our project and defined the methodology that we will use. We have two options for…

1

Week 3 — Like I Like This week we decided our datasets which we can use in our project and defined the methodology that we will use. We have two options for dataset; Paris and Oxford. We will start with using the Paris one but later we can change to Oxford one too. We will use a neural network algorithm and we will try to apply feed-forward neural network. DATASETS Sample photo from Paris Dataset [1] The Paris dataset is divided into some sections: La Defense (381),Eiffel Tower (147),Hotel des Invalides (452),Louvre (268),Moulin Rouge (445),Musee d’Orsay (740),Notre Dame (445),Pantheon (662)Pompidou,(441)Sacre Coeur (380)Arc de Triomphe (554) and general Paris photos (1497). Sample Photo From Oxford Dataset [2] The Oxford dataset, which has the same features as the Paris dataset, contains 16 buildings images and general images of Oxford. There are 5062 images in total. METHODOLOGY Feed Forward neural network is basically bunch of neurons that allows information flow only in one direction. Information from one input unit pass to the next layer of unit and after they compute something, they pass their output to the next unit and this goes like that until you get to very end. They do not contain any cycles or loops and there is no connection between perceptrons in the same layer. Sample of feed-forward neural network This algorithm works to reduce the error rate which is the difference between output and desired value d. The value of error tells us how far away we are from the desired value for a particular input, therefore the algorithm tries to minimize it. We have just started the coding part. We will see and share the changes on project on the following days! References: [1] J. P. a. A. Zisserman. ”the paris dataset, 2008.http://www.robots.ox.ac.uk/ vgg/data/parisbuildings/. [2] R. A. a. A. Z. James Philbin. The oxford buildings dataset. http://www.robots.ox.ac.uk/ vgg/data/oxbuildings/.

Week 3 — Like I Like

0

week-3-like-i-like-1252958e3bf0

2018-05-11

2018-05-11 03:56:09

https://medium.com/s/story/week-3-like-i-like-1252958e3bf0

false

295

Course Projects for Introduction to Machine Learning, an undergraduate class at Hacettepe University — This semester the theme is Machine Learning and The City..

null

bbm406f17

null

bbm406f17

MACHINE LEARNING

null

Machine Learning

machine-learning

Machine Learning

51,320

Like I Like

null

151e80cd261a

likeilike

3

1

20,181,104

null

0

null

0

null

2018-02-13

2018-02-13 02:28:54

2018-02-13

2018-02-13 02:32:58

1

false

en

2018-02-13

2018-02-13 02:32:58

0

1252ad67d0de

0.716981

1

0

Decision tables are used often in businesses to model business rules. They are such an important tool as they may describe different…

1

IS 3500 — Assignment 3.1 Decision tables are used often in businesses to model business rules. They are such an important tool as they may describe different outcomes due to different decisions that may be made by either an employee or a user of the system. This tool is also useful in both testing and requirements management as it allows users to visually see the outcomes from specific decisions. A decision table consists of decisions as well as outcomes. In the example provided, a customer can be a VIP Gold, Silver, or no VIP customer at all. Based on their status and the amount that they spend in the store on an annual basis, the customer is entitled with a voucher than can be used for store purchases. The decision table can be modeled as such below.

IS 3500 — Assignment 3.1

1

is-3500-assignment-3-1-1252ad67d0de

2018-02-13

2018-02-13 13:24:09

https://medium.com/s/story/is-3500-assignment-3-1-1252ad67d0de

false

137

null

Data Science

data-science

Data Science

33,617

Jonathan Blaustein

2nd Year. Information Science. Northeastern University.

87803256892d

blaustein.jo

2

20,181,104

null

0

null

0

null

2018-09-12

2018-09-12 16:10:50

2018-08-10

2018-08-10 15:37:33

2

false

en

2018-09-12

2018-09-12 16:13:21

7

125521e8c797

4.036164

0

Welcome back to Fintech Love Island. Earlier this week we flew to sun-caressed Mallorca to check out our 10 Fintech Trends competing to be…

5

Fintech Love Island — Introducing our Finalists Welcome back to Fintech Love Island. Earlier this week we flew to sun-caressed Mallorca to check out our 10 Fintech Trends competing to be the hottest of the hot: AI, Gamification, Mobile Wallet, Blockchain, Roboadvisory, Chatbot, API, VR, Robotics and P2P. Today we are holed up, along with our 10 Fintech Trends, in the lounge of the Fintech Villa, beset with all manner of fans and water-coolers lest the heat further sap our decision-making powers. Now let’s get one thing straight — all ten of these Fintech Trends are magnificent specimens, as well-toned in heart and spirit as they are in body. However, every dream summer holiday must come to end, and Fintech Love Island is no exception. It’s time to give our avid, zombified viewers exactly what they’ve been waiting for: a winning Fintech Couple. In our previous post we determined, rigorously and beyond all reasonable doubt, that Mobile Wallet, Roboadvisory, Chatbot, VR and P2P are… not hot. Or at least not as hot as the other Trends. As these five notties are bundled off the premises, we turn to our 5 remaining Trends: AI Gamification Blockchain API Robotics Surviving acronyms AI and API exchange a furtive look of complicity; robotics remains impassable as ever, as impervious to pressure as he is to joy; and in the background, the whirring of a fan. The near-silence is broken by the jingle of a mobile phone. Someone forces their way through the thronged camera crews with a handwritten note for the judges. The producers, it appears, have a strict No Threesome Policy and are insisting that the judging proceed on a couple-by-couple basis. A memo from the Audience Insights Manager is attached as well, pointing out that the gross preponderance of acronyms within Fintech engenders great bewilderment and confusion, and that this could be responsible for the low levels of engagement detected in certain key audience segments. API has got the shakes all of a sudden and keeps trying to catch AI’s eye — but AI is intent on staring straight forwards and will not be deterred. Application Programming Interface versus Artificial Intelligence … there could only really be one winner. The heavies are waiting in the wings and, at the nod of the judges, descend on poor API, who, kicking and screaming, vanishes through a side door. This leaves us with two Fintech Couples: AI & Blockchain on the one hand, and Gamification & Robotics on the other. Let’s size them up: Couple #1: AI & Blockchain** AI is no spring chicken — but, after many years of flighty bachelordom (or the AI Winter as those in the know tend to call it), he finally appears ready for something more substantial. And people having been whispering for a while that 2018 could be Blockchain’s year. We may finally see commercialised use cases outside of cryptocurrencies, where she is undisputed queen. Couple #2: Robotics & Gamification** Robotics is possibly our most promiscuous Trend. Whenever there’s a business process, which, ahem, needs improving, you can count on him being somewhere close by, sniffing around with that robot nose of his. As for Gamification, well, the name says it all really. Whether we’re talking gameplay, challenges, targets or continuous mobile engagement, consumers can’t get enough of her in 2018. ** We would like to add that the construction of our Trends as heterosexual couples is for parodistic purposes only, rather than as a statement of gender norms. At buzzvault, we have a firm commitment to inclusion irrespective of gender or sexual orientation. Our two Fintech Couples will now leave the Fintech Villa to spend the weekend in their respective Fintech Chalets. The chalet concept was something we came up with as a means of giving our Trends a more peaceful, secluded environment where they can get to know each other a bit better whilst remaining under uninterrupted 24-hour CCTV surveillance. So, AI & Blockchain rise from their sofa and, hand-in-hand, walk down the steps of the Fintech Villa, at the bottom of which a tastefully upholstered limo awaits them. Their chalet sits on the magnificent Playa del Cangrejo Enajenado on the south coast, where under the full moon turtles crawl onto the shore for their once-yearly breeding. As AI & Blockchain are whisked away to their private beach, Robotics & Gamification are heading to the helipad round the back of the Fintech Villa. Their chalet is in the hills, up on the East face of the picturesque Montaña del Gran Armiño. Awaiting them once they arrive is a selection of locally — and sustainably — sourced tapas. In our two follow-up posts, we’ll check in on each of our Fintech Couples to see how they’re getting on. As with any relationship, what we’re looking for is a complementary fulfilment of needs — our winning couple must be greater than the sum of its parts. Join us next week then, where — first up — we’ll be catching up with Robotics & Gamification. Fintech power couple or match made in hell? There’s only one way to find out … Forwards to next post: let’s see how Robotics and Gamification are getting on … About buzzvault Our mission is simple: helping customers protect the things they love with personalised home & contents insurance. Leveraging our world-first digital asset vault on the blockchain, we make it easy for customers to securely catalogue their possessions from the convenience of their mobiles, giving them cover matched to their needs. If you’d like to be part of our journey to fix home and contents insurance, please sign up here for buzzvault beta. We’d love to know what you think. Originally published at blog.gobuzzvault.com on August 10, 2018.

Fintech Love Island — Introducing our Finalists

0

fintech-love-island-introducing-our-finalists-125521e8c797

2018-09-12

2018-09-12 16:13:21

https://medium.com/s/story/fintech-love-island-introducing-our-finalists-125521e8c797

false

968

null

Fintech

fintech

Fintech

38,568

buzzvault 🏠

buzzvault is a secure digital asset vault that lets customers digitally catalogue their possessions in an app. Download the app and get your quote today. Easy.

8031070e3920

buzzvaultHQ

0

1

20,181,104

null

0

न्यूरल नेटवर्क के बुनियादी सिद्धांत स्नातकोत्तर शोधकर्ताओं के लिए एक व्यापक पुस्तक है। पुस्तक बताती है कि उन्नत वास्तविक दुनिया समस्याओं से निपटने के लिए तंत्रिका नेटवर्क कैसे उपयोग किए जा सकते हैं। यह सभी प्रमुख तंत्रिका नेटवर्कों पर चर्चा करता है, जिसमें बताया गया है कि उनके आर्किटेक्चर अलग-अलग कैसे हैं। पुस्तक में एल्गोरिदम भी प्रस्तुत करता है, जो कि स्वयं के कार्यक्रमों में तंत्रिका नेटवर्क को लागू करने के लिए प्रोग्राम में रूपांतरित किया जा सकता है। यह पुस्तक सभी प्रोग्रामरों और छात्रों को अपने स्वयं के काम में तंत्रिका नेटवर्क को लागू करने के लिए एक अनिवार्य संसाधन है। Laurene Fausett के बारे में Lauren वी। Fausett एक गणितज्ञ और अकादमिक है दक्षिण कैरोलिना Aiken विश्वविद्यालय में गणित के एक प्रोफेसर के रूप में काम कर रहे उसने कई सरकारी वित्त पोषित परियोजनाओं जैसे कि भूमिगत कोयला गैसीकरण, बाधित क्षेत्र में भिगोना, चैलेंजर आपदा के बाद अंतरिक्ष शटल पर ओ-रिंग का विश्लेषण, और कृत्रिम तंत्रिका नेटवर्क के अनुप्रयोगों के लिए काम किया है। कैलिफोर्निया विश्वविद्यालय, कैलिफोर्निया के एक स्नातक, डॉ। फॉएब्स ने वायोमिंग लैरमी विश्वविद्यालय, वायोमिंग में अपनी मास्टर और डॉक्टरेट की डिग्री का पीछा किया। उन्होंने यह भी लिखा है: निर्देशक के मैनुअल फॉर फाउंडान्मेंटल्स ऑफ न्यूरल नेटवर्क्स, नमूना सॉल्यूशंस एंड सॉफ्टवेयर, एप्लाइड न्युमेरिकल एनालिसिस मैटलैब का प्रयोग करते हैं, और न्यूमेरिकल एनालिसिस विद एप्लीकेशन एण्ड एल्गोरिदम।

1

null

2018-01-12

2018-01-12 17:13:16

2018-01-12

2018-01-12 17:17:56

1

false

en

2018-01-12

2018-01-12 17:17:56

1

125585efc6eb

2.049057

0

FAUSETT’s Fundamentals of Neural Networks: Architectures, Algorithms and Applications, 1e

2

Fundamentals of Neural Networks: Architectures, Algorithms and Applications, 1e FAUSETT’s Fundamentals of Neural Networks: Architectures, Algorithms and Applications, 1e Fundamentals of Neural Networks is a comprehensive book for postgraduate researchers. The book explains how Neural Networks can be used to tackle advanced real world problems. It discusses all major neural networks, explaining how their architectures differ. The book also presents algorithms which can be adapted into programs to implement neural networks into one’s own programs. The book is an indispensable resource for all programmers and students trying to implement neural networks into their own work. About Laurene Fausett Lauren V. Fausett is a mathematician and academician working as a Professor of Mathematics at the University of South Carolina Aiken. She has also worked for many government funded projects such as underground coal gasification, constrained layer damping, analysis of the O-rings on the space shuttles after the Challenger disaster, and applications of artificial neural networks. A graduate of the University of California Berkeley, California, Dr. Fausett pursued her Master’s and doctoral degrees at the University of Wyoming Laramie, Wyoming. She has also written: Instructor’s Manual for Fundamentals of Neural Networks, Sample Solutions and Software, Applied Numerical Analysis using Matlab, and Numerical Analysis with Applications and Algorithms. Fundamentals of Neural Networks: Architectures, Algorithms and Applications, 1e IdeaKart.com | Buy Fundamentals of Neural Networks: Architectures, Algorithms and Applications, 1e book online at best…www.ideakart.com Binding Paperback Language English Number Of Pages 480 Author FAUSETT Publisher Pearson Education India Isbn-10 8131700534 Isbn-13 9788131700532 And for those who love Hindi and want to learn in Hindi this book is about:

Fundamentals of Neural Networks: Architectures, Algorithms and Applications, 1e

0

fundamentals-of-neural-networks-architectures-algorithms-and-applications-1e-125585efc6eb

2018-05-11

2018-05-11 04:45:34

https://medium.com/s/story/fundamentals-of-neural-networks-architectures-algorithms-and-applications-1e-125585efc6eb

false

490

null

Machine Learning

machine-learning

Machine Learning

51,320

Ideakart

Providing you best things at best prices!

2cd96c165111

ideakart24

1

2

20,181,104

null

0

null

0

null

2018-06-04

2018-06-04 23:37:38

2018-06-05

2018-06-05 00:41:30

5

false

zh-Hant

2018-06-05

2018-06-05 00:41:59

6

125669fffc5f

1.112579

0

今日主題: Gradient Boosting Model

2

Day 15 — Gradient Boosting Model (GBM) 今日主題: Gradient Boosting Model 有一些中文翻譯直翻提升方法，大概跟估狗翻譯小姐的翻譯功力差不多。但我沒好到哪去，根本不知道怎麼翻，就用英文原文吧。參考資料 Friedman, Jerome H. “Greedy function approximation: a gradient boosting machine.” Annals of statistics (2001): 1189–1232. Prince Grover Wikipedia Towards Data Science itw01 Complete Guide to Parameter Tuning in Gradient Boosting (GBM) in Python 這次的主題相對中文尤其正體中文的資料少了很多。大概能找到寫得比較好又是專門針對GBM的就是這些了。 [1]就不用多提，GBM的創始論文，5525次引用。 [2]講了許多關於Gradient Boosting的基礎概念。並不專講GBM但是把數學理論簡單介紹了一下。 [3]連中文翻譯頁面都沒有，大概還真的是沒人去翻譯吧? [4]有點像是[1]的講解，看不懂或沒時間看[1]的話就看這個就夠了。 [5]是少數找到完整的中文資料，果然用母語學東西快得多。如果不想看其他的引用資料其實光看[5]就足夠懂個大概。 [6]偏向python實作的文章。裡面把各個要調整的參數都列了出來，實戰上調整參數的重要性就不用我多說，透過這篇文章去了解需要調整哪些參數是很有幫助的資源。心得 GBM的核心觀念是為輸入資料建立M種不同的分類/迴歸模型。這M種模型都是比較簡單的弱分類器(Weak Learner)。每次分類都將上一次分類錯誤的資料提高權重再分類一次。最後得到的M個模型就會有M種權重，最後的分類結論就是將這M個模型按照權重加總起來就是答案了。數學公式可以下圖來解說參考[5]，對每個模型m，都對前m-1個模型求梯度，得到最快下降的方向，就是這第m個模型的梯度。演算法流程如下分支之一: Least-squares regression 分支之二: LAD TreeBoost 分支之三: M TreeBoost 原文還有其他分支以及評估方法，就得回去查了。 Code: 參考[6]的範例

Day 15 — Gradient Boosting Model (GBM)

0

day-15-gradient-boosting-model-gbm-125669fffc5f

2018-06-05

2018-06-05 00:42:00

https://medium.com/s/story/day-15-gradient-boosting-model-gbm-125669fffc5f

false

74

null

Machine Learning

machine-learning

Machine Learning

51,320

Falconives

null

250d8013fad2

falconives

11

19

20,181,104

null

0

null

0

null

2018-01-29

2018-01-29 14:40:57

2018-01-29

2018-01-29 14:52:01

1

false

ru

2018-01-30

2018-01-30 08:50:07

3

12569bec7206

2.479245

1

0

В надежде нарастить продажи украинские ритейлеры все чаще используют рекомендации на основе алгоритмов машинного обучения. Потребителям…

5

О тонком искусстве покупательских рекомендаций В надежде нарастить продажи украинские ритейлеры все чаще используют рекомендации на основе алгоритмов машинного обучения. Потребителям затея эта нравится, но лишь тогда, когда ритейлер в состоянии «считать» их личные предпочтения и интересы. А до этого в Украине все еще далеко. image source: http://mallenbaker.net Недавно с одним моим знакомым приключилась забавная история. Известный интернет-магазин rozetka.ua проанализировал его историю покупок и не нашел ничего оригинальнее, чем предложить «сделку дня» — наушники по цене 39,000 грн. (или примерно 1370 дол. по текущему курсу). С товаром они угадали, с ценой — сильно рассмешили. Нет, мой знакомый действительно тратит на покупки через rozetka.ua приличные суммы, однако не в таких масштабах. О чем он собственно и написал в Facebook. К чести онлайн-ритейлера минут через пять после постинга сотрудники rozetka.ua прислали еще одно предложение с более «земной» ценой (раз в десять ниже). Впрочем тут, подозреваю, не обошлось без деликатного вмешательства пиар-службы, своевременно «отловившей» проблемный отклик в сети. Конечно, отладка алгоритмов — дело кропотливое и затратное. Требует серьезных инвестиций как в людей, так и в оборудование & софт. К примеру, в 2006 году компания Netflix ради всего лишь 10%-ного улучшения своих рекомендательных алгоритмов устроила конкурс с призовым фондом в миллион долларов, затянувшийся на несколько лет. (Редко какой учебник по машинному обучению сейчас обходится без упоминаний об этом конкурсе. Хотя, на мой взгляд, куда интереснее послушать его участника, Chris Volinski, Assistant Vice President for Big Data Research из AT&T Labs в недавно запущенном подкасте DataFramed). Наши родные ритейлеры таких гигантских призов не предлагают. Вот еще, зачем?)) Вместе с тем, заметно, что вопросами найма аналитиков они с некоторых пор озадачились. Правда, по старой привычке все еще надеются переманить кадры у конкурентов или же взять на работу неоперившихся студентов. Как бы там ни было, вакансии аналитиков с неизменно размытой формулировкой «умение работать с большими массивами данных» теперь не редкость. В принципе, текст вакансии и его место размещения довольно много могут рассказать о самом работодателе. К примеру, если вакансии проявляются в основном на досках объявлений, а не на корпоративных сайтах, то стоит засомневаться в приоритетах компаний. В конце концов что для них важнее: прощупать рынок на предмет зарплаты, чтоб, не дай бог, не переплатить, взяв первого встречного, или действительно максимально «закрыть» проблемный участок работы. Если в вакансии не конкретизированы инструменты для работы «с большими массивами данных», автор текста, по всей видимости, крайне слабо ориентируется в теме. Ведь проанализировать наполнение покупательской корзины (market basket analysis), с одной стороны, можно с помощью ПО таких «тяжеловесов» как Oracle, SAP или, на худой конец, с помощью комбинации последних версий MS Excel+SQL Server. С другой, для тех же целей существуют весьма продвинутые open source инструменты типа R-модулей arules, arulesViz и uplift, а также scikit-learn и XGBoost (работают под Python). Однако наши эйчары такими деталями почему-то не злоупотребляют. А жаль, ибо упоминание в CV правильной должности+предыдущего работодателя вовсе не гарантирует, что они «автоматом» получат человека, умеющего писать хорошо отлаженные алгоритмы, после чего все покупатели неистово полюбят именно этого ритейлера. Все-таки товарная номенклатура может серьезно отличаться от магазина к магазину (даже не по части категорий, а по наполнению покупок — не везде ведь покупают сразу несколько единиц). К тому же, не секрет, что в Украине звучные должности редко подразумевают интересный насыщенный функционал и хороший соцпакет. Куда чаще приходится выдавать «на гора» фиксирующие очевидное отчеты, которые потом мало кто, если вообще читает. В результате получаем торговые сети с завышенными ценами и странным ассортиментом, а также ироничные постинги юзеров в соцсетях обо всех подобных «проколах». Хорошо еще, когда пиар-служба успеет отмониторить эти отзывы и погасить скандал в начальной стадии. Но пиарщики — не волшебники. Без аналитической поддержки коллег долго в таком режиме не протянут.

О тонком искусстве покупательских рекомендаций

1

о-тонком-искусстве-покупательских-рекомендаций-12569bec7206

2018-02-01

2018-02-01 19:21:37

https://medium.com/s/story/о-тонком-искусстве-покупательских-рекомендаций-12569bec7206

false

604

null

Algorithms

algorithms

Algorithms

7,319

Alla Khrystych

Analyst, big believer in the power of SQL, Python and R. Intrigued by ideas at the intersection of marketing, tech and finance. Views are my own.

27e8230f1c98

alla_khrystych

112

180

20,181,104

null

0

null

0

413a65ad8247

2017-09-23

2017-09-23 09:51:05

2017-09-23

2017-09-23 10:02:31

3

false

en

2017-10-02

2017-10-02 19:57:57

1

1256c97292d5

2.999057

3

0

I am going to automate or remove any manual task in my life that doesn’t provide me value by doing it manually.

3

Automating Life I am going to automate or remove any manual task in my life that doesn’t provide me value by doing it manually. “All jobs will be automated in the next 50 years…” OK (add sarcastic tone) There is a future of autonomous self-driving, crypto cashless payments, predicted shopping orders, off the grid power with a quick trip to Mars for a weekend break. We are experiencing parts of this today in varying degrees of complexity and reward, but the truth is no one knows which way the future will turn out (that is the cool thing). By 80s predictions, we should all be in silver suits and zooming around space like the Jetsons. The one thing that has remained constant in history is human desire to automate, driven by greater goals or simply increased profits. Either way, automation is a continuing force in our lives, spanning generations, no one is sitting with an abacus at his or her desk, and the office calculator is a thing for only a few. As humans, we move with the times. The technology revolution has begun, and though we believe it to be in full swing, personally I feel it has not even started. Automation of once manual tasks will accelerate with exponential growth over the coming decades, and though will seem slow on a macro basis, we will see a rapid change in the major of job roles in our lifetime. So why try and automate in your life? Three factors (Now, Near, Future) Now — Life is busy (I do cause this though), there are many things I prefer to do and believe I can gain time to do them now by automating those lack value or add negativity. Near — Doing many tasks you dislike isn’t good for you mentally or physically, I want to protect myself in the near term from mental and physical fatigue. Future — Automation is happening, I want to get mentally prepared for the future, both through learning and enjoying the extremes of free-time we may live to enjoy. This series is not a “How to automate your life in 3 months” series, but a mindset shift for the next 30–40 years. I do aim to automate the least valuable and mundane tasks of my life in the next 6–12 months. So what is the point… There are 8,760 hours in a year (not that many when you read it like that), and I want to get value from as many of them as possible. Sharing between work, health, fun, family, self, and activities. That is 168 hours per week. Currently, I need a solid 7.5 hours of sleep a night, so that leaves 115.5 hours per week (16.5 hours per day) to optimise. Below are some of the areas I am focusing on first, this shift is not about cutting a journey time from 20 minutes to 15 minutes but looking to see if that journey does not provide value or is negative to creating value. Such as changing the timing of that journey, so I am not in stop and go traffic which causes frustrating and not adding any value to my week. Key areas to start with: fitness, travel (commutes/business trips and holidays), household chores, sleep/rest, paperwork, food shopping, family coordination, consuming content, business meetings, family time and writing. Why document? I have always been a fan of automation in business and personally and want to share and collaborating with others as feel this will be something we all ask ourselves at some point. “Are the things I am doing providing me value.” If your thinking about giving it a try, just do it and please let me know how you got on. you can email me via leemallon.com if I can help in any away. Subscribe for updates on how I am automating my life via www.leemallon.com

Automating Life

5

automating-life-1256c97292d5

2018-06-08

2018-06-08 23:01:52

https://medium.com/s/story/automating-life-1256c97292d5

false

649

Sharing learnings

null

leemallon

lee@rarely.io

leemallon

XAMARIN FORMS,AUTOMATION,BOXING,ALEXA,TECHNOLOGY

leemallon

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Lee Mallon

Just trying to figure it all out... Apart of @rarelyhq

b36ebcc27864

leemallon

352

452

20,181,104

null

0

null

0

null

2018-08-18

2018-08-18 16:24:19

2018-09-05

2018-09-05 04:02:24

4

false

en

2018-09-05

2018-09-05 05:01:48

2

1256d846d2ec

4.001887

0

Description:

4

Deal or no Deal — Predicting Lending Club Loan Outcomes Description: This project was completed for my 3rd project as part of the 12 week Metis Data Science immersive bootcamp. The goal of this project was to use machine learning classification algorithms to accurately predict classes for a dataset. Project Design: Like any other data science project, the first step is to obtain a data set that fits the needed criteria. I decided to use the 2015 loan data from Lending Club, a peer to peer money lending platform for loans between $1000 and $40,000. This dataset contained only accepted loans. I decided to use this dataset because I see a lot of useful applicability in learning to comprehend financial datasets and the types of information they contain. This dataset was also extremely robust in terms of number of features and number of usable records, which is something I wanted to get more experience in. I decided to focus on predicting whether a loan will be “Fully Paid” or “Charged Off”, making this a binary classification analysis. After selecting the predictor, I had to determine what metric I am trying to optimize for with my classification. With this in mind, my proposed business case was to: Optimize for precision of ‘Fully Paid’ loans to predict which loans would be “low risk” with the goal of maximizing return on investment for a lender. I chose precision because in this business case, the cost of a false positive (Predicted: Fully Paid, Actual: Charged Off) is high for a money lender. Tools Pandas Matplotlib GridSearchCV Sklearn Seaborn Train/test split Cross Validation Random oversampler Were a few of the packages that I utilized on this project for analysis and visualization. I had experience with most of the tools by this point in the bootcamp, but this project exposed me further and significantly increased my comfort level in running different modeling algorithms. Data As mentioned above, the dataset I used was 2015 LendingClub.com accepted loans. This was a large dataset with originally more than 500k observations. After I decided to do a binary classification, I was able to reduce the dataset down to 300k usable observations by eliminating some of the other outcomes (such as loans that were still current). Below are a few of the useful fields and their corresponding datatype Class Imbalance Within the dataset, there was a big class imbalance. The classification I was attempting to predict, ‘Fully Paid’, was a much more frequent class than ‘Charged Off’, as seen below. Random Sample of 10,000 records from the dataset With this in mind, I decided to utilize sklearn’s random oversampling technique to balance out the classes. This allowed me to compensate for the lack of observations for the ‘Charged Off’ class, allowing my model to generalize better overall. Feature Selection My strategy for attacking feature selection went along these lines: Debt to Income ratio (DTI) was determined to be the most important feature by the random forests algorithm Eliminate all redundant features (ex: Accounts opened in last 6 months vs accounts opened in last 12 months) Eliminate features that were unfairly descriptive (ex: Amount of loan that has already been paid off). These features would be unfair to use on the premise that the model is meant to be used at the genesis of a loan, rather than after time has already passed. Utilized Random Forests and the XGB Classifiers feature importance quantification. This allowed for an empirical approach to feature selection. After eliminating features based on this criteria, I was down to roughly 63 features. Algorithms One of my goals for this project was to understand and utilize a variety of different algorithms and the use cases for them. The table below describes all of the algorithms I used: Algorithms used on the dataset The dummy classifier, which just predicts the majority class for every prediction, preformed quite well because we are predicting the majority outcome with ‘Fully Paid’. Going up from the dummy, we can see that most of the models did improve on the baseline. Using an exhaustive GridSearchCV search with a multitude of different hyper-parameters, I was able to obtain a precision of 92% on my test set using the support vector machine algorithm, which shows that my model does generalize quite well on new data. Below is a confusion matrix of my highest performing model. We can see here that my model is predicting 2,155 of the 2,334 observations correctly. Conclusion Overall, my model did end up generalizing quite well on the dataset. Looking at the important features, it was no surprise that debt to income (or dti) ratio was one of the most telling features for a potential borrower. I was, however, surprised that LendingClub’s given loan “grade” was not a very significant predictor for a loans outcome. Besides E and F graded loans (the lowest), there are a significant amount of loans paid back that might not be graded the best. Since this project was for a bootcamp, I was constrained by my deadlines and need to continue onto my next project. If I were to take this project further, this is some of the future work I would be interested in pursuing: Non-binary classifier (by grade, etc) Feature Engineering Deploy flask app in production Experimentation with different class imbalance methods Please reach out if you have any questions or comments! You can find all my code over here.

Deal or no Deal — Predicting Lending Club Loan Outcomes

0

deal-or-no-deal-predicting-lending-club-loan-outcomes-1256d846d2ec

2018-09-05

2018-09-05 05:01:48

https://medium.com/s/story/deal-or-no-deal-predicting-lending-club-loan-outcomes-1256d846d2ec

false

875

null

Machine Learning

machine-learning

Machine Learning

51,320

Harmeet Hora

Industrial Engineer Turned Data Scientist

ed938c00d257

harmeethora

3

20,181,104

null

0

null

0

4fb63eafb690

2018-07-31

2018-07-31 15:16:50

2018-08-08

2018-08-08 08:14:30

6

false

en

2018-08-21

2018-08-21 11:30:55

7

125b26ef737e

5.059434

16

0

From an industrial problem to a cutting-edge quality control solution

3

The story behind the AI-driven startup Deevio From an industrial problem to a cutting-edge quality control solution Introducing Deevio Ever wondered how startups come to life? Let me introduce you to Deevio and the story of how we went from idea to reality. Deevio comes from the selective innovation process of the Berlin-based deep tech company builder WATTx — a specialized, cross-functional team that is built to create the next wave of industrial startups. At the core is a user-centric approach to ideation and new business creation where only the best ideas survive. So, let´s go six months back in time and see the process behind building the next generation of industrial quality control. The idea: deep learning for image classification The project started when we were approached with the idea of using deep learning for image classification in the fields of industrial quality control (machine vision). We were keen to give the idea a try, knowing the potential of the technology in manufacturing. Experimenting with the tech A week later, we received the first two data sets to work with and it quickly became clear that we had something special here. We rapidly achieved accuracy results of close to 100% with low resolution images using state-of-the-art data science methods. When presenting these results to experts from the machine vision industry, they did not believe what they saw — as the prevailing opinion is that the higher resolution you have, the better accuracy you will get. With our approach, that was no longer the case. This was the first indication for us that there in fact was great potential in the idea. Summary of the first training results — high accuracy on downscaled images Key takeaway: you don’t need expensive cameras, good algorithms get good results even on low-resolution images. But wait: is there an actual problem to solve? The first results looked promising, however, just having a great technology without a clear problem to solve is never a good start for a startup. It is something we at WATTx try to avoid at all costs by doing extensive research before committing to new projects, both from the user experience and the business side. Quality control in manufacturing is often still done manually So, before investigating the technology further, we spent time better understanding the process of quality control and visual inspection in the manufacturing space. We soon found out that quality control in manufacturing is often still done manually as traditional, rules-based machine vision systems fail to address a lot of use cases. This manual work is exhausting for the operators doing it and prone to errors as it is impossible for humans to detect every small defect of a product through an 8-hour shift. We also found out that even though machine vision technology has been around for years, rules-based systems’ shortcomings limit a widespread usage of the technology. Especially the long setup time, the lack of flexibility and the high costs are limitations that came up regularly in the interviews. Key takeaway: deep learning gives the flexibility needed to make machine vision a practical solution, as well as, massive cost benefits. Is there a market for it? Besides the strong interest we got from our interview partners and the potential of AI applications in factories, the spending on quality control is going from one record high to another. With this last piece of validation, we went on to a final sense check with the WATTx team to discuss whether to pursue the project or not. In the meantime, we did another validation of what we can do with Deep Neural Nets: classifying a set of nails we bought at the hardware store next door Key takeaway: When we overcome the limitations of machine vision systems, we believe that we can significantly increase the market demand for it. Finally, Do or Die? The goal of do-or-die sessions is to kill the project by finding reasons not to pursue with the project in order to detect all possible shortcomings. This way we ensure that only the most promising ideas survive and eventually become independent companies. Since we had a clear problem to solve, strong results on the tech side, positive feedback from industry experts and first and foremost a team that was highly passionate about the project, we decided to bring this project to life. The prototype: from nails to an actual case Now, we had to tackle one of the hardest tasks when starting a startup — coming up with a name. After lots of meetings, brainstormings and ideation sessions, we came up with Deevio. Deevio.ai - a name combining both worlds, artificial intelligence and machine vision We continued to get a lot of positive feedback and ideas for use cases for Deevio from industry experts and soon these conversations turned into discussions on pilot projects. These became our first feasibility studies. Parallel to working on the new data sets from industry partners, we experimented with different hardware platforms, industrial cameras and ways to integrate our solution to PLCs (these guys basically send the signals on production lines and are important for automating processes in factories) to build our first working prototype. Our first prototype: a machine vision camera, semi-professional lighting, an Nvidia Jetson TX2 running our deep learning model for classifying the nails and an interface showing the output Really understand the user needs Then it was time to step up the UX research and design: we travelled throughout Germany to visit various factories and talk to the people working there. In this way, we ensured that we designed a solution that is best suited to the end-user needs. We got feedback from both line and plant managers as well as quality assurance representatives and used it to design the first interfaces. One major finding was to keep the interface for the operators as simple and clear as possible to avoid any uncertainties. Thus, instead of adding lots of features in the beginning, our designer came up with the most minimalistic interface he ever designed. The Deevio team at the factory of our partner Viessmann with the mission to better understand product operators and identify potential application areas The prototyping phase ended with a confirmed customer, ongoing feasibility studies, various other promising pilot projects in the pipeline and a lot of new insights from seeing all these factories and talking to users. The MVP: launching Deevio We are now launching Deevio with the mission of augmenting and helping the people that carry out visual inspections and providing real-time insights in the production lines for manufacturers. If you want to know more about Deevio and our journey, stay tuned. Soon we will tell you more on what it’s all about, how we designed the customer interaction and how exactly we are supporting operators in their manual inspection tasks using deep learning. *** If you’re interested in working with us or if you’re already having a use case from your production environment in mind, feel free to reach out to me directly at damian@deevio.ai or via our website.

The story behind the AI-driven startup Deevio

213

the-story-behind-the-ai-driven-startup-deevio-125b26ef737e

2018-08-21

2018-08-21 11:30:55

https://medium.com/s/story/the-story-behind-the-ai-driven-startup-deevio-125b26ef737e

false

1,089

Read the freshest news about the machine vision industry and get insights on how to use deep learning to improve industrial quality control

null

deep vision

contact@deevio.ai

deep-vision

MACHINE VISION,DEEP LEARNING,QUALITY CONTROL,ARTIFICIAL INTELLIGENCE,INDUSTRY

DeevioBerlin

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Damian Heimel

Co-founder at deevio.ai, transforming industrial quality control with deep learning

d07c5c79968d

damian.heimel

14

20,181,104

null

0

null

0

863f502aede2

2018-05-11

2018-05-11 20:14:55

2018-05-11

2018-05-11 20:15:54

3

false

en

2018-05-11

2018-05-11 21:21:12

2

125d6725ad13

1.776415

4

0

Beijing-based artificial intelligence startup Unisound, (a.k.a. Yunzhisheng / 云知声) today announced that it had raised US$100 million in…

5

China’s Voice Tech AI Startup Unisound Raises US$100 Million to Boost IoT Services Beijing-based artificial intelligence startup Unisound, (a.k.a. Yunzhisheng / 云知声) today announced that it had raised US$100 million in Series C funding, the richest-ever single funding round for a smart voice technology startup. The investment was led by CLP Health Fund, joined by 360 Technology, Qianhai Wutong M&A Fund, Hanfor Capital, etc. Chinese media also reported that the company has already started its agenda-setting Series C+ funding. Founded in 2012, Unisound aims at making Internet of Things (IoT) devices smarter by adding AI capabilities such as voice recognition, language processing, knowledge computing and big data solutions. In a 2016 interview with Synced, Unisound Founder and CEO Huang Wei laid out his company’s ambitious “Chip+Edge+Cloud” strategy for producing processors and sensors to empower edge devices, enabling voice-activated interactions for users, and connecting edge devices with Unisound’s cloud-based intelligent platform. Unisound puts a particular emphasis on smart home appliances, vehicles, and healthcare. Last month, Unisound and Shanghai-based Phicomm jointly launched the smart speaker R1, which leverages cutting-edge voice recognition and semantic analysis techniques to better understand human speech. The company accounts for more than 70 percent of the aftermarket share in China’s automotive field, and its voice interaction solutions for healthcare are deployed in more than 60 major Chinese hospitals. Unisound’s supercomputer platform has reached a benchmark performance of 10 petaflops. Last year, Unisound received CNY¥300 million (US$45 million) in investments for development of its first AI chip, UniOne, which will be released on May 16. * * * Journalist: Tony Peng| Editor: Michael Sarazen * * * Subscribe here to get insightful tech news, reviews and analysis! * * * The ATEC Artificial Intelligence Competition is a fintech algorithm competition hosted by Ant Financial for top global data algorithm developers. It focuses on highly-important industry-level fintech issues, provides a prize pool worth millions. Register now!

China’s Voice Tech AI Startup Unisound Raises US$100 Million to Boost IoT Services

26

chinas-voice-tech-ai-startup-unisound-raises-us-100-million-to-boost-iot-services-125d6725ad13

2018-05-14

2018-05-14 04:26:59

https://medium.com/s/story/chinas-voice-tech-ai-startup-unisound-raises-us-100-million-to-boost-iot-services-125d6725ad13

false

325

We produce professional, authoritative, and thought-provoking content relating to artificial intelligence, machine intelligence, emerging technologies and industrial insights.

null

SyncedGlobal

null

SyncedReview

global.sns@jiqizhixin.com

syncedreview

ARTIFICIAL INTELLIGENCE,MACHINE INTELLIGENCE,MACHINE LEARNING,ROBOTICS,SELF DRIVING CARS

Synced_Global

Startup

startup

Startup

331,914

Synced

AI Technology & Industry Review - www.syncedreview.com || www.jiqizhixin.com || Subscribe: http://goo.gl/Q4cP3B

960feca52112

Synced

8,138

15

20,181,104

null

0

null

0

null

2018-03-19

2018-03-19 11:30:00

2018-03-19

2018-03-19 11:31:49

1

false

en

2018-03-19

2018-03-19 11:31:49

7

125dbaa4eebf

4.033962

2

0

These travel spots around the world are showcasing latest technological innovations in Virtual Reality, Artificial Intelligence and…

5

Top Travel Destinations Any Tech Lover’s Bucket List Should Include These travel spots around the world are showcasing latest technological innovations in Virtual Reality, Artificial Intelligence and Robotics Tired of the same-old trips? Why not take a tech-themed trip and explore some of the latest — and arguably strangest — tech out there? We selected some destinations where technology is the main attraction. And if you have any further ones to add to that list we’d love to hear from you! Virtual Reality Theme Park There is a lot of hype around Virtual Reality (VR) and these days anybody can get a taste of the sort of immersive experience that this technology offers by using relatively affordable devices. From the likes of Samsung GearVR which retails at around $100 to Google Cardboard — which was given out for free with promotional campaigns such as those run by the New York Times and The Guardian. Yet to get the full impact of the futuristic potential of VR, some argue that you actually need to travel in the real world, to a place called The Void. Simply put, the Void is a Holodeck-like theme park that is the ultimate in VR immersive entertainment. Its main location is in Lindon, Utah, but it also recently opened peripheral experiences in Dubai and in New York’s Madame Tussauds (where it put visitors into a Ghostbusters-themed narrative complete with proton guns). “The best VR uses more than just your eyes,” says Void Founder Curtis Hickman, who was part of the team originally looking to build a theme part called Evermore in Salt Lake City but pivoted to the emerging field of Virtual Reality after the park was eventually put on hiatus. He believes that fully interactive VR such as they offer will play a big role in the development and growth of the industry going forward. The idea behind it is that in order to be truly taken into this alternative virtual world, you need to engage all your senses, so the experiences at The Void are all finely tuned with effects — such as heat, wind and sounds — to match the visual narrative as you move through a real-life course. A lightweight motion-tracking vest also provides haptic feedback such as vibrations that are triggered when you’re — for example — attacked by a ghost. For those who fancy the idea of fully merging the real and virtual world, that’s pretty hard to beat. Artificial Intelligence Concierges Hotels that have historically struggled to keep up with competition from tech-centric competitors such as Airbnb now seem to be striking back on the technology front. As Artificial Intelligence becomes smarter and more pervasive, many hotels are, for example, using AI-powered concierges to enhance their guest experience. Hilton Worldwide’s AI concierge, Connie (named after company founder Conrad Hilton), was arguably the first true AI-powered concierge bot, powered by IBM Watson and travel database WayBlazer it advised guests of local attractions and interesting sites, fine-tuning its responses based on frequent requests. In Las Vegas’ Cosmopolitan, their own AI assistant Rose entices guests with a calling card which is handed to them upon check-in which reads, “Know my secrets. Text me,” and “I am the answer to the question that you never asked.” Unlike Connie, however, Rose has a phone number which guests can text at any time, making it a much more flexible service as you don’t need to physically find her. At the Novotel München Messe, Munich Guests are also welcomed by a virtual concierge. There are impressively large touch screens distributed around the hotel, offering accessible information on local attractions, weather and flight information. Guests can even send a virtual postcard at the tap of a button. Guests staying at Radisson Blu, meanwhile, can request the assistance of AI chat bot “Edward” by text message. The group stated in a press release that Edward is “designed to deliver exceptional experiences for guests who prefer digital brand interaction”. It can deliver information on local bars and restaurants as well as deal with complaints. For problems and questions it is unable to handle, though, human staff is still on hand to help. Robot Hotels Starwood rolled out two robotic “Botlrs” named A.L.O. in their Cupertino Aloft Hotel back in 2015 and they’ve been a huge hit with the techy crowd ever since. The robotic butlers, built by Savioke, are able to perform tasks in front and back of house, as well as navigate around guests and use elevators using a combination of sensors and WiFi/4G connectivity to communicate with the hotel and elevator software. The bots deliver amenities — such as a toothbrush or extra towels — to guest rooms in lieu of actual humans. When the robot arrives at the room, the guest can enter in a rating on the robot’s touchscreen, or offer a “tip” in the form of a tweet to the hashtag #MeetBotlr. According Brian McGuinness of Aloft Hotels, using robots frees up staff time so they can create a more personalized experience for guests, adding that there were plans for implementing the bots in other locations if the pilot continued to prove successful. In Japan, however, you can take that love of robots to the next level by staying at the Henn-na Hotel. Housed in a Netherlands-themed amusement park (why not?) it is staffed entirely by robots. This includes a life-sized velociraptor — which wears a periwinkle bow tie, a bellhop cap, and a neckbeard — used for checking in English-speaking patrons, a process that typically takes only about five minutes, much shorter than average check-in times experienced in human-staffed hotels. The hotel had 72 rooms and only 10 human staff. Those staff members were present for emergencies, should they arise, and cleaning. In the future, the hotel hopes to replace even these with cleaning robots (there are already robots employed for cleaning carpets in addition to robotic luggage porters and personal assistants). And if anybody can pull that off, it’s probably Japan.

Top Travel Destinations Any Tech Lover’s Bucket List Should Include

2

top-travel-destinations-any-tech-lovers-bucket-list-should-include-125dbaa4eebf

2018-03-19

2018-03-19 13:41:53

https://medium.com/s/story/top-travel-destinations-any-tech-lovers-bucket-list-should-include-125dbaa4eebf

false

1,016

null

Virtual Reality

virtual-reality

Virtual Reality

30,193

PredictX

PredictX makes the invisible visible. By bringing trends, anomalies, and patterns to the surface, you can better identify growth and savings opportunities.

804c52824b12

predictx

1

4

20,181,104

null

0

null

0

null

2018-08-15

2018-08-15 12:53:24

2018-08-15

2018-08-15 13:01:26

1

false

en

2018-08-15

2018-08-15 13:01:26

2

125eb010aa8a

1.871698

0

Shaloo Garg is the Head of Oracle’s Innovation and Global Startup Ecosystem, and a Global Advisor on Technology and Innovation for UN…

3

60 Seconds with WITS Southeast Keynote Speaker — Shaloo Garg Shaloo Garg is the Head of Oracle’s Innovation and Global Startup Ecosystem, and a Global Advisor on Technology and Innovation for UN Women. Her keynote talk at WITS Southeast is “The Next Digital Frontier- Impact of AI & Machine Learning on Global Economy, Society and Philanthropy”. Why is speaking at WITS important to you? Diversity across all spectrums is an important component of a constructive conversation. WITS brings an audience from various verticals, levels of management and solution areas. Over the years, I have seen WITS grow and bring converstions to the table which are pertinent for change and growth for women and girls in our society. What inspires you? The strongest use case of technology is it’s use in the social impact space. What inspires me every single day is how technology is changing the world around us. What drives me every single day is to uncover use cases where conservative ways of working can be done away with the help of technology especially in under developed, developing and under served communities. Most useful article you have read in the last one month Want An Innovative Business? Start With Gender Equality Want an innovative business? HP's Nate Hurst says that to do this you need to start with advancing gender equality…www.forbes.com What did you want to be when you were a kid? My dad was an aeronautical engineer and we often travelled with the family. I noticed that the air hostess in the airplane had a selfless job. They would serve food, water to the passengers with dedication. I was totally fascinated by that. For the longest time, I dreamt of being an air hostess :) Blurb about your session The Next Digital Frontier- Impact of AI & Machine Learning on Global Economy, Society and Philanthropy Innovations in digitization, analytics, artificial intelligence (AI), and machine learning are creating performance and productivity opportunities for business and the economy, even as they reshape our daily lives and future of work. The wave of AI and Machine Learning is coming at us very rapidly. In fact, it has already permeated parts of our professional and personal lives. In this power packed session, learn about the impact of AI and Machine Learning on the Global economy, society and philanthropy. Let’s also talk about channels of disruption that will be the front runners in this race and these will be the ones that will be the game changers ! Whether we are want it or not- are you prepared to capture value from the oncoming wave of innovation of AI and Machine Learning? Let’s find out !

60 Seconds with WITS Southeast Keynote Speaker — Shaloo Garg

0

60-seconds-with-wits-southeast-keynote-speaker-shaloo-garg-125eb010aa8a

2018-08-15

2018-08-15 13:01:27

https://medium.com/s/story/60-seconds-with-wits-southeast-keynote-speaker-shaloo-garg-125eb010aa8a

false

443

null

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Women In Tech Summit

The Women In Tech Summit inspires, educates and connects women in the technology industry. We support @techgirlzorg and all women in tech.

cb7754decea5

WomenTechSummit

851

1,499

20,181,104

null

0

null

0

null

2018-06-15

2018-06-15 11:16:34

2018-06-15

2018-06-15 11:24:44

1

false

en

2018-06-15

2018-06-15 11:35:04

1

125eed2acb4d

1.939623

1

0

Crypto trading and bots fit together like the moon and the earth. you can’t say crypto without tradingbots. but why?

5

5 reasons why you should opt for a Bot Crypto trading and bots fit together like the moon and the earth. you can’t say crypto without tradingbots. but why? 1. They’ve got a proven track record As of 2014, 75% of all trades taking place on the New York Stock Exchange occurred at the hands of automated trading systems (ATS). Why? Because ever since they were introduced, they proved to be much better traders than people are. Naturally the early ATS adopters were the real winners because they were competing with a market controlled my humans. Cryptocurrency trading is much newer than stock trading, but funnily enough it has still not entered the age of automated trading, until now. 2. They’re complete workaholics Let me put it this way. Humans eat, sleep and live their life. Cooks cook, bankers bank, pilots fly. Of all of us day traders and would-be-cryptocurrency-connoisseurs, how many of you can really say you dedicate your entire life to trading? 0. Trading bots literally work all day, all night. Always, always, always scanning price fluctuation, finding opportunities and avoiding catastrophe. 3. They don’t have emotions Sure, you’re better at writing a love poem to your crush, but when it comes to trading, the bots are just better because they’re 100% rational. Ever panic sold when you thought a coin was about to tank and then watched it boom back up minutes later? If you answered yes, you’re most likely a human. Most major stock exchanges have trading curbs that close markets for some time to counteract panic selling. This ensures traders to have the time and ability to assess accurate information and think rationally…kind of like a bot would. 4. They are YOU Not so fast, I’m not saying you’re the terminator. The point here is that your bot doesn’t work until you set it up. It’s up to you to tweak and adjust its decisions to trade exactly how YOU would if you had the time, energy and muscle to stay glued to the charts all day. The working bot is therefore a reflection of you and your trading style. 5. They’re easy to use Ok, ok, they’re great at trading — even better than humans, but that also means they must be complicated…WRONG. You do not need a Phd from MIT to control a personal trading companion. These things have been designed to be easy to set up and start out. Better yet, if you don’t know anything about trading, there are plenty of highly capable and successful trader more than willing to share the best configurations across social media platforms like youtube, Facebook groups, discord etc. Happy Hopping! Cryptohopper.com Cryptohopper is an automated cryptocurrency, that runs 24/7 in the cloud. https://www.cryptohopper.com/?atid=4531

5 reasons why you should opt for a Bot

1

5-reasons-why-you-should-opt-for-a-bot-125eed2acb4d

2018-06-15

2018-06-15 11:35:05

https://medium.com/s/story/5-reasons-why-you-should-opt-for-a-bot-125eed2acb4d

false

461

null

Trading

trading

Trading

19,801

Cryptohopper

null

dbf162f35f87

cryptohopper

196

17

20,181,104

null

0

null

0

null

2018-03-21

2018-03-21 14:54:30

2018-03-21

2018-03-21 14:57:26

0

false

en

2018-03-21

2018-03-21 14:57:26

0

125f8592b9e1

0.030189

0

null

5

Become a friend of Destiney at www.destiney-ai.com

0

become-a-friend-of-destiney-at-www-destiney-ai-com-125f8592b9e1

2018-03-21

2018-03-21 14:57:27

https://medium.com/s/story/become-a-friend-of-destiney-at-www-destiney-ai-com-125f8592b9e1

false

8

null

Money

money

Money

35,618

DESTINEY VISION

null

760d85bbbc65

destineysocialnetwork

2

1

20,181,104

null

0

null

0

e0de41c7612d

2018-02-19

2018-02-19 17:11:11

2018-02-19

2018-02-19 21:17:40

0

false

en

2018-02-19

2018-02-19 21:17:40

1

1260241e2a3e

2.769811

1

0

My attempts with online learning resources.

2

MOOCs My attempts with online learning resources. Inspired by Tirthajyoti Sarkar’s post on choosing effective courses for machine learning and data science, I’m going to take stock of my progress so far and make an effort to figure out how I can better use the available educational resources. Pre-2018 Relevant Courses On evaluating the classes that I’ve taken, out of 21 audited MOOCs, I wouldn’t classify any of the contents covered at beyond a basic level of comprehension. This is my subjective view of course, but it gets at one of the biggest issues that I have with MOOCs: I can’t seem to move forward. MOOCs have been great for me to understand a breadth of material but I haven’t been able to get deeper into the subjects that interest me. (Data Science | Basic) Data Visualization and Communication with Tableau, Duke University, Coursera (Data Science | Basic) The Data Scientist’s Toolbox, Johns Hopkins University, Coursera (Data Science | Basic) Getting and Cleaning Data, Johns Hopkins University, Coursera (Data Science | Basic) Regression Models, Johns Hopkins University, Coursera (Data Science | Basic) A Crash Course in Data Science, Johns Hopkins University, Coursera (Data Science | Basic) Developing Data Products, Johns Hopkins University, Coursera (Data Science | Basic) Building a Data Science Team, Johns Hopkins University, Coursera (Data Science | Basic) Data Science in Real Life, Johns Hopkins University, Coursera (Data Science | Basic) Managing Data Analysis, Johns Hopkins University, Coursera (Data Science | Basic) Computing for Data Analysis, Johns Hopkins University, Coursera (Data Science | Basic) Introduction to R for Data Science, Microsoft, edX (Data Science | Basic+) Introduction to Recommender Systems, University of Minnesota, Coursera (Data Science | Basic+) Practical Predictive Analytics: Models and Methods, University of Washington, Coursera (Finance | Basic) Competitive Strategy, Ludwig-Maximilians-Universität München, Coursera (Finance | Basic) Analyzing Global Trends for Business and Society, University of Pennsylvania Wharton School of the University of Pennsylvania, edX (Programming (py) | Basic) An Introduction to Interactive Programming in Python (Part 1), Rice University, Coursera (Programming (py) | Basic) Programming for Everybody (Python), University of Michigan, Coursera (Programming (Web) | Basic+) HTML, CSS, and Javascript for Web Developers, Johns Hopkins University, Coursera (Programming (Web) | Basic) HTML, CSS and JavaScript, The Hong Kong University of Science and Technology, Coursera (Programming (Web) | Basic) Responsive Website Basics: Code with HTML, CSS, and JavaScript, University of London International Programmes Goldsmiths, University of London, Coursera (Programming (Web) | Basic) HTML5 Coding Essentials and Best Practices, World Wide Web Consortium (W3C), edX After making this list (which took a while), I can see that at least part of the cause of my problem is that I keep picking basic courses to take. Reflecting upon how I picked these courses, I think the sequence of events unfolded something like the below: The MOOC platforms started to offer Specializations I did not keep track of which Specialization the course I took belonged to or the future classes in the Specialization When I finished a course and searched for another class on my topic of interest, I came upon more Specializations I looked at the syllabus for the first course and there would always be at least one item that I didn’t feel like I had a good grasp of The result is that I keep taking the first few courses in multiple Specializations It’s surprisingly hard to keep track of courses when only auditing (this is probably intentional). Also, the lack of structure in the self-paced classes hasn’t worked so well for me either. But since I’m getting all of this top rated education available for no cost, I can’t complain and it is really on me to keep track of my learning. At this point now with so many people sharing their own experience of self learning data science, it shouldn’t be too hard to come up with a curriculum as a student. Colleges also publish their data science curriculums, and using that as a basis might be a good idea. Another point that I should consider is that most of the classes I’ve taken have been through Coursera, and Coursera restricting the quizzes and assignments has been detrimental to my learning. I should probably explore more non-Coursera options too. So this is my big takeaway: keep track of courses and specializations and create a curriculum of courses that can build upon one another.

MOOCs

2

moocs-1260241e2a3e

2018-04-04

2018-04-04 17:34:08

https://medium.com/s/story/moocs-1260241e2a3e

false

734

Stories and reflections on understanding our imminent overlords. The theory and practice of using statistics and computing algorithms.

null

Human in a Machine World

null

human-in-a-machine-world

DATA SCIENCE,MACHINE LEARNING

null

Data Science

data-science

Data Science

33,617

JJ

Thinker and Tinkerer. Figuring out a storyline. http://yahwes.github.io/

5086e44692d2

captioned

173

16

20,181,104

null

0

null

0

null

2018-06-14

2018-06-14 01:45:17

2018-06-14

2018-06-14 02:41:12

1

false

en

2018-06-14

2018-06-14 02:43:50

7

12604dbf1e56

1.075472

0

Let me start by providing some context.

5

How to make money on Coinbase’s GDAX Photo by Andre Francois on Unsplash Let me start by providing some context. I’ve just watched the presentation video by Ambika Sukla on “Automatic financial econometrics with AI”. Ambika, Executive Director of AI in Morgan Stanley, presented how to apply deep learning algorithms to accurately represent the price trend, to infer volatility and to predict when to buy or sell. What captivated me is how he showed that Convolutional Neural Network, commonly applied in computer vision problems, can be used to accurately represent price trends. Then he showed how he used Generative Adversarial Network in inferring price volatility. Lastly, he showed how Reinforcement Learning can be used to implement a fully automatic virtual technical trader which knows when to buy or sell. The point I inferred from the presentation is that it is possible to use Deep Learning algorithms, without applying any sophisticated technical analysis, to make profit on GDAX. Of course, the caveat is that I’m assuming that you are a Python programmer and you have a basic understanding of deep learning algorithms. The series of stories which I will publish will cover the following: Deep Technical Trader Architecture Overview (late June) Getting GDAX price data (early July) Training and Validating deep learning models (late July) Using Deep Technical Trader on GDAX (mid August) Through the series, you will be able to implement a fully automatic technical trader. Stay tuned!

How to make money on Coinbase’s GDAX

0

how-to-make-money-on-coinbases-gdax-12604dbf1e56

2018-06-14

2018-06-14 02:45:41

https://medium.com/s/story/how-to-make-money-on-coinbases-gdax-12604dbf1e56

false

232

null

Machine Learning

machine-learning

Machine Learning

51,320

Ivan Tang

null

54482b21fd08

hiimivantang

20

20,181,104

null

0

null

0

4e218b28b6c2

2017-11-05

2017-11-05 10:40:12

2017-11-06

2017-11-06 21:25:40

11

false

en

2018-08-14

2018-08-14 17:32:06

4

1260726bfebd

3.922642

20

1

0

Discovering artwork with visual search, or the reasonable effectiveness of convolutional neural networks

5

Art Genius Discovering artwork with visual search, or the reasonable effectiveness of convolutional neural networks This article was initially published November 6th, 2017, prior to the acquisition of Thread Genius by Sotheby’s. Try out the demo at art.threadgenius.co Something we’ve always wanted to try was expanding the Thread Genius visual search experience to include discovering artworks based on visual similarity. It was never highly prioritized because training models for a new domain usually means gathering entirely new training data. In the same way we trained our fashion neural net to learn about silhouettes of dresses and the different variations of plaid, it would make sense that we’d have to do the same for aspects of art. But out of curiosity, how well would our existing models do on art? That is, would the visual features learned from fashion apply to discerning similarities amongst pieces of art as well? In this blog post, we explore this question further. Composition vs. Subject matter At Thread Genius, we have a few neural net models trained to recognize various concepts found in imagery. One model, nicknamed FashNet or Fashion Model, was trained only to recognize concepts related to fashion. These include patterns (stripes vs. camouflage), shape (dress vs. pants), colors (blue vs. red), and embellishments (buckles vs. epaulettes). Another model, which we call internally as Super Model, was trained to recognize fashion concepts as well as general concepts that most people would know about — think animals, plants, buildings, etc. This was trained primarily to minimize false positives when dealing with user-generated photos. Without introducing any new concepts about art, we created two search indices by running these models on a catalog of 600K+ artworks. Comparison of search results from two different models: one trained on only fashion images (i.e. “Fashion Model”), and the other trained simultaneously on both fashion and general real-world objects (i.e. “Super Model”). You can see a trade-off between a focus on texture and color (composition) vs. what the piece is about (subject matter). A comparison of the results from these two models shows that there’s a trade-off. Our fashion model doesn’t know anything about things outside of fashion. What are clouds? Never seen them before but they look like feather prints. Apples? Nah, those are probably red watches. So when you task it to extract visual features from images about concepts it doesn’t know, there are interesting effects. One is that there’s an emphasis on what it does know, things like colors and textures. To the Fashion Model, an oil painting of an apple is just a round red thing with blotchy textures, and so it’ll group all round, red, blotchy things together. To the Super Model, since it knows about apples, it’ll group apples together. Fashion Model is great for abstract art, Super Model less so. Super Model is great for sculptures, Fashion Model less so. Depending on your taste, you may prefer one over the other. Side note: Interestingly, the fashion model found an apple painting that appeared to be the same exact one as the input image. On closer inspection, they’re actually two different paintings from two different artists: the input image is from Jane Palmer and the search result is a 2016 piece from George Cassallo. OOOOOH. Search results from Fashion Model One thing we find fascinating is that, although we never trained these models to know anything about artists or painting techniques, pieces by the same artist naturally get grouped together based on similarities in brush strokes, color choice, etc. Degrees of Separation Of course, what would a deep learning blog post be without a giant t-SNE image of our embeddings? Obligatory money shot follows. We used to have this demo at Spotify called “Boil the Frog,” in which we grabbed two random songs and used some machine learning techniques to find a chain of songs that gradually morphed one into the other. Here are some examples of this concept applied to works of art. Examples of art laddering Art laddering, animated GIF-style More examples of art ladders What’s next? Last month we launched our API which allows any developer to access our visual search engine. In fact, developers on the API have access to the same fashion model that we used to produce the results in this blog post. Admittedly though, getting our models to work well for all types of artwork would require some additional fine-tuning. You saw how there’s a trade-off between a model that emphasizes composition vs. a model that emphasizes subject matter. Finding an optimum model would mean finding the right balance between these two strengths.

Art Genius

257

art-genius-1260726bfebd

2018-08-16

2018-08-16 17:16:58

https://medium.com/s/story/art-genius-1260726bfebd

false

695

Supporting the future of art and technology

null

sothebys

null

Sotheby's

null

sothebys

ART,ARTIFICIAL INTELLIGENCE,DATA SCIENCE,DESIGN,LUXURY

Sothebys

Artificial Intelligence

artificial-intelligence

Artificial Intelligence

66,154

Thread Genius

We’re mapping the world’s visual taste

48d331020f0e

ThreadGenius

319

17

20,181,104

null

0

null

0

855952a2eb7d

2018-08-04

2018-08-04 17:50:59

2018-08-17

2018-08-17 14:26:52

10

false

zh-Hant

2018-08-17

2018-08-17 14:26:52

3

126102e87b9b

1.642453

1

0

Evaluate Model Performance with Loss Function

4

R語言自學日記(15) -評估模型表現 Evaluate Model Performance with Loss Function “Two people in elegant shirts brainstorming over a sheet of paper near two laptops” by Helloquence on Unsplash 前言還記得我們在AR(p)模型裡面學到的評估模型優劣的方式，也就是AIC或是BIC準則，這些準則的概念是，透過最小化模型預測出來的結果與實際結果之間的誤差，來選擇最佳的模型。 Source：Wikipedia 上圖是一個簡單的迴歸模型，藍色的線是我們預測出來的結果，紅色的點是實際的資料分布，而綠色的線我們稱之為誤差，可想而知，一個好的模型應該具有最少的誤差，也因此我們創建了一個方式來衡量這些誤差的總值，我們稱作損失函數(Loss Function) 歐幾里得距離與統計距離 (Euclidean Distance and Statistical Distance) 在開始講損失函數之前，我們要先談距離的觀念，在誤差的衡量上有很多時候都是以預測點跟實際資料點之間的最小距離來做衡量的，頂多有些模型因為樣本、變數數量等因素會加上懲罰項。而最小距離的概念是甚麼？我們這邊要介紹兩個概念：歐幾里得距離與統計距離。歐幾里得距離是被定義在n維歐幾里得空間中的距離，這個空間是可以被擴展到任意維度的實內積空間，聽起來很複雜，但如果我們讓n = 3，其實就是我們生活中所習慣的三維空間，當然二維空間、一維空間也是我們所熟悉的。而對於資料點的世界來說，每多出一筆獨立資料我們就可以說整個資料集的世界多出了一個維度，以下我透過程式模擬一個二維資料集與一個三維資料集來看看：二維空間與三維空間資料模擬：Iris Dataset 在歐幾里得空間中的距離概念非常簡單，兩點之間的最小距離可以透過以下的式子衡量，也就是直線距離：然而，在統計資料的世界中，資料點並不一定分布在歐幾里得空間(Non-Euclidean Space)裡面，造成這個現象的原因有很多，比方說變數的相關性，或是變數的變異數差異，都會造成非歐幾里得空間的分布狀況，我們可以想像成是空間被扭曲了，舉例來說像是莫比烏斯帶：莫比烏斯帶示意圖；Source：Wikipedia 上圖就是一個比較典型的非歐幾里得空間，這種設計我們很常看到，跟我們直覺不同的是，如果你站在莫比烏斯帶的任一個點上往前走，基本上你是可以無限走下去的，你很難想像他並非一個一維或是二維空間。因此，要衡量兩個點之間的距離，就並不一定是透過歐幾里得空間裡定義的最小直線距離，比方像是機器學習中的流形學習(Manifold Learning)，對於點的計算方法就非常不同，這些方法雖然複雜，但也有助於我們更加貼近真實世界，比方說用譜嵌入(Spectral Embedding)去實現人臉辨識技術。一般常見的統計距離有很多種，舉例來說我們可以透過除掉標準差來表示出標準化歐氏距離(Standardized Euclidean Distance)，或是透過共變異數矩陣去清除相關性，我們稱為馬氏距離(Mahalanobis Distance)：實務上我們不見得能夠完全了解資料的分布狀況，尤其是對於高維度的大數據而言，但我們可以用各種損失函數去評估誤差，這等同是一種後驗方法，即使我們不知道實際距離狀況，卻可以模擬出該距離計算下的總誤差，也就某種程度上去反推得資料的大致分布情況。常用的損失函數：MAE、MSE/RMSE 我們可以透過不同的損失函數來評估一個模型，在這邊先介紹兩個最常用的，MAE(平均絕對誤差，Mean Absolute Error)與MSE(均方誤差)。我們先透過剛才使用過的鳶尾花資料來建立一個Linear Regression模型：接下來我們可以分別去Fit MAE與MSE損失函數看看，我們先看一下兩個函數分別的計算公式：如果把MSE去做根號運算，就變成RMSE(Root Mean Square Error)，兩個函數的實際解釋是，MAE是計算每個點到Y=X上的誤差，而RMSE計算的是每個點到所有資料平均值的誤差。在一般的情況下MAE與RMSE得到的結果是非常相似的，但是隨著資料出現更大的變異數或是離群值，RMSE通常會比較大，這給了我們一個觀念是，RMSE放了更大的權重在Error Magnitude，但本身不見得對於平均誤差有足夠的解釋力。延伸閱讀： MAE and RMSE — Which Metric is Better? Mean Absolute Error versus Root Mean Squared Errormedium.com 我們透過代碼來看一下這個迴歸模型的誤差：要特別注意的是，我們不可以使用不同的變數準則去決定一個模型的優劣，應該要建立在同一準則下不斷修正誤差，比如調整權重、放入參數等，舉例來說我們可以改成多項式迴歸或是多變量迴歸，去檢驗單一指標的總誤差，最後選定最優值來建構估計模型，但是不可以使用同一個模型而只選擇好的估計指標。如何預測：Training、Validation & Testing 時間序列資料的驗證通常需要經過時間(廢話)，如果我們建立了一個預測10期，一期一個月的ARIMA模型，至少要等到10個月後才能評估該模型的總誤差，等於說為了測試哪個模型預測能力最好，我們通常要等10個月。因此，在第七章短暫提過的 Training Set 與 Testing Set 就會派上用場，測試資料通常要等10個月，因此實際上我們會把訓練集資料再切分成Training Set與Validation Set，也就是我們假裝不知道一段期間內的資料變化，而透過其他已知的資料去建構模型，來比較模型預測結果與被我們假裝忽略的資料的總誤差，以便於評估這個模型的效力，稱為樣本外預測(Pseudo out-of-sample forecasting)。下一個難題是，如何選擇資料期間？單一的資料期間在時間序列資料上很容易出問題，我們很難以確保趨勢、季節性與結構性都是一致的，此時，我們通常會使用不同的方法做更精準的估計，我們在這邊介紹其中一種方法也就是遞迴法(Recursive Scheme)，如果以一個AR(1)模型舉例： Source：陳旭昇，時間序列分析 R是指樣本內資料，P則是指樣本外資料，由上述的式子可以知道，我們的估計式會不斷變化，樣本數也會不斷增多，我們可以不斷的把時間往後推，對於1到6月的資料，我們拿7月驗證，對於2到7月的資料，我們拿8月驗證等。 Overfitting & Underfitting 最後一節我們要談建立模型中常遇到的兩個問題：過度配適(Overfitting)與低度配適(Underfitting)。過度配適：當我們在訓練集上得到的結果很理想(比如誤差很小，或R平方很高)，但是在測試集上得到的結果很差的時候，通常就是發生過度配適的問題，要解決這樣的問題有許多辦法：減少變數的使用：這非常好理解的是，當我們要辨識一個人的善惡，只要我們能夠有他從小到大所有的資料，這個判斷大概都會變得很準確，但如果拿同樣的標準去套用在其他人身上，就不見得準確，對於機器學習來說也是一樣，在解釋能力不要差異太大的情況下，應減少預測變數。調整模型參數：比較常見的例子像是，在一個L1懲罰項作為損失函數的模型評估中，使用的模型卻是以L2參數來配適的。這個步驟非常複雜，也常常是資料科學家的重頭戲之一。(Kaggle上的競賽幾乎一定都會有暴力調參的環節) 交叉驗證(Cross-Validation)：簡單來說就是重複切割資料，並且對每次切割都配適一次模型，如果發現每次配適的結果差異很大的時候，通常就是有過度配適的問題了。此外，通常經過交叉驗證的模型雖然誤差會比較不理想(每一次Train的樣本更少)，但是結果通常更可靠。整體學習(Ensemble Learning)：這邊的方法有很多，之後有空再一一介紹，比方說將基層模型(Base Model)做疊加，或是用不同的抽樣方法(Boostrap Sampling)等等，概念上是透過連續、重複隨機等不同方法去盡可能讓機器學習到整體的資料以避免偏誤。重新研究你的輸入資料，是不是變數取捨或特徵工程沒有做好等等低度配適：當我們在訓練集上得到的結果不理想，但是測試集上的結果較優(實際上這個問題比較複雜，但不排除是Underfitting的情況)，有可能就是模型低度配適，在小樣本的預測上比大樣本更好一些，顯現模型的泛用性並不是這麼高，通常可以用加大變數、交叉驗證等方式解決。總結下一篇我們會談到如何去檢測時間序列資料裡面的異常值與結構性變動，這可以幫助我們針對不同的區段分別建構預測模型，以更加精準的預測，如果你喜歡我的文章，還請下方不吝按下Clap喔！

R語言自學日記(15) -評估模型表現

11

r語言自學日記-15-評估模型表現-126102e87b9b

2018-08-17

2018-08-17 14:26:53

https://medium.com/s/story/r語言自學日記-15-評估模型表現-126102e87b9b

false

104

About a self-taught diary on R Language programming and practical Time Series Analysis, made by a python user and BBA student. Hope you like it:)

null

R 語言自學系列

poiuy8568@gmail.com

r-語言自學系列

DATA SCIENCE,TIMESERIES,R LANGUAGE,SELF TAUGHT,DATA ANALYSIS

null

Data Science

data-science

Data Science

33,617

Edward Tung

Senior BBA Student, Data Science, Consulting Intern

b8b9ba7ac6eb

poiuy8568

69

21

20,181,104

null

0

null

0

1770da37c3a6

2018-05-24

2018-05-24 11:51:45

2018-05-24

2018-05-24 12:00:58

3

false

en

2018-05-24

2018-05-24 12:03:43

2

1263b77af2db

4.546226

3

1

0

It’s been a busy year for the team here at OSA DC, and developments have especially picked up over the past few months.

5

1st Place Prizes, Warm Receptions, and My Talk with Ban Ki Moon: An Update on OSA DC’s Progress Alex Isaev and Ban Ki Moon It’s been a busy year for the team here at OSA DC, and developments have especially picked up over the past few months. We’ve been showcasing our project at conferences all around the world to get the word out on the revolutionary work we are undertaking. Since the beginning of the year, we’ve attended 15 conferences in 10 countries across 3 continents. At the d10e conference in Seoul, South Korea, we had the honor of being awarded the 1st place prize for best tech startup. Each conferences was a success in its own right, but as the 1st place prize at d10e suggests, our ventures into South Korea were especially fruitful. I personally had the privilege to attend the Synco Blockchain Conference in Seoul at the end of March, and the overall feedback and reception OSA DC received was nothing shy of glowing. All of our ideas — the current available product with the OSA Hybrid Platform and the future product we’re working towards with OSA DC — were very well received. The project itself is vast and ambitious, so most observers initially responded to it by saying, “Yes, this is a great idea, but it’s too big — how do you plan to implement it?” Little did they know, the OSA Hybrid Platform is already being used by companies like Coca Cola, Danone, and MARS and that our business model is already profitable. When we told them this, they were shocked. “Wow,” they’d say. “You’ve built it and you’re already making money from it? That’s huge!” Of course, our the OSA Hybrid Platform is largely for business to business enterprise solutions. Business from all over the world, be it China, Japan, Korea, Russia, the United States, are very receptive of our model and welcome its solutions. But at Synco Blockchain and other conferences and meetups, we introduced OSA DC and its business to consumer solutions to a wider audience. Many attendees in these audiences were potential contributors — your every day individuals who could benefit from transparent product information and smart shopping solutions. When we introduced the OSA DC product, the consumer’s role on the platform, and how it will make shopping easier and safer for retail customers, they were thrilled. Everywhere we went, those who listened to our pitch showed great enthusiasm for the project and its potential. Some — and this happened more often than not — even stood up during the demonstration and said things like, “Thank you! I’ve had enough with retailers and their practices. I have two kids and I don’t know what groceries to buy; I don’t know what is in the food I’m purchasing because I don’t trust the product information. This is exactly what we need.” OSA DC will certainly help businesses to cut back on waste and increase their bottom lines. But behind these solutions, the platform’s business to consumer tools will make shopping a safer and more efficient experience for retail customers all over the world — with OSA DC, we’re ushering in the era of the smart consumer, helping individuals stay healthy and save money with our AI and blockchain-based Data Master Catalog. It was wonderful to connect with so many businesses and consumers who resonated with OSA DC’s mission. There was one encounter in particular during my trip to Korea, however, that was especially fruitful: my meeting with former UN Secretary-General, Ban Ki Moon. Lunch with Ban Ki Moon I was invited to a lunch with Ban Ki Moon, and we discuss the impact of blockchain on legacy businesses and the widespread issues of global hunger and the tens of billions of dollars in annual food waste. Obviously, this is where OSA DC came into the conversation; we talked about the benefits blockchain and AI solutions could bring to the retail industry, and I also extended OSA DC’s AI application to your average household, wherein this artificial intelligence could track and mitigate product waste within each home. We call this reliable AI, which focuses on minimizing product waste, not helping businesses increase profits. We then discussed how OSA DC could help consumers make healthier decisions. With our Data Master Catalog, customers can crosscheck any product’s information/nutrition facts to find the healthiest foods possible; and all of this information is secured by the immutability of the blockchain’s distributed ledger, meaning it is fully transparent and can’t be altered. All of these concepts were well received by Ban Ki Moon, and he said he is interested in tracking the project’s growth. By leaguing ourselves up with prominent political figures like Ban Ki Moon, we have left the doors open for cooperation with governments, leading scientists, and other public sector entities in the future, all of which can help us achieve our dream and deliver retail solutions that will benefit the entire human race. As you can see, it’s truly a busy and exciting time for all of us here at OSA DC. The reception we received at international conferences and Ban Ki Moon’s interest in the project demonstrate that the fruits of our labor are flowering forth into a project that is quickly growing into its potential. Like a bloom in the Spring day’s sun, OSA DC is truly blossoming. As we continue into the next phases of the project, our ICO, product development, and our expansion into Asian and American markets, we’re excited to see the impact of our solutions take shape. We’re confident that, once people see how our platform revolutionizes product information verification and shopping services, OSA DC will make waves in the retail industry. With OSA DC, smart consumers will have their interests protected to lead healthier lives, all while businesses cut back on costly product waste. The future is bright, and we’re looking to it with great optimism. Together, we can work together to change retail for the better — one transparent product at a time. Avanti con amore! Alex Isaiev - Co-Founder, Business Development Lead - OSA decentralized (OSA DC) | LinkedIn View Alex Isaiev's profile on LinkedIn, the world's largest professional community. Alex has 8 jobs listed on their…www.linkedin.com Ban Ki-moon - Secretary General - United Nations | LinkedIn View Ban Ki-moon's profile on LinkedIn, the world's largest professional community. Ban has 1 job listed on their…www.linkedin.com

1st Place Prizes, Warm Receptions, and My Talk with Ban Ki Moon: An Update on OSA DC’s Progress

53

1st-place-prizes-warm-receptions-and-my-talk-with-ban-ki-moon-an-update-on-osa-dcs-progress-1263b77af2db

2018-06-13

2018-06-13 22:44:44

https://medium.com/s/story/1st-place-prizes-warm-receptions-and-my-talk-with-ban-ki-moon-an-update-on-osa-dcs-progress-1263b77af2db

false

1,059

OSA DC is a decentralized, AI-driven blockchain platform that collects and analyzes data from retailers, manufacturers, consumers, and open data sources in real-time.

null

OSAdecentralized

null

osadc

null

osadc

null

global_OSADC

Smart Consumer

smart-consumer

Smart Consumer

5

Alex Isaiev

Smart consumer. CEO, OSA DC.

1ae447a3d0cc

alexisaiev

89

81

20,181,104

null

0

null

0

null

2018-05-31

2018-05-31 04:23:42

2018-06-23

2018-06-23 13:34:16

21

false

th

2018-06-30

2018-06-30 03:41:43

22

1264028567a5

7.162264

13

0

ระดับ : Beginner / Intermediate / Advanced เชิง : Concept / Mathematics / Implementation / Creativity / Others แท็ก : Deep Learning…

4

มารู้จักอักขราอาร์เอ็นเอ็น โดยลองเล่นแต่งเป็นกลอนอักษรสาร ระดับ : Beginner / Intermediate / Advanced เชิง : Concept / Mathematics / Implementation / Creativity / Others แท็ก : Deep Learning, Natural Language Processing, Language Modeling เกริ่นนำ Karpathy Andrej Karpathy [ภาพจาก Medium] ผู้ที่อยู่ในแวดวง machine learning คงจะคุ้นเคยกับชื่อนี้กันเป็นอย่างดี เพราะเมื่อไม่กี่ปีก่อน Karpathy นับเป็นดาวรุ่งที่พุ่งแรงที่สุดคนนึง ซึ่งสรรค์สร้างผลงานต่างๆ ไว้มากมาย ไม่ว่าจะเป็นการสอนคอร์ส CS231n ที่เปรียบเสมือนการเปิดศักราชคอร์ส deep learning สมัยใหม่ รวมทั้งยังเขียนเปเปอร์ เขียนโปรแกรม และเขียนบทความหลายชิ้น อันล้วนแล้วแต่น่าตื่นตาตื่นใจและเป็นหมุดหมายสำคัญของวงการ ในที่นี้เราจะกล่าวถึงบทความนึงของ Karpathy ที่มีชื่อว่า “The Unreasonable Effectiveness of Recurrent Neural Networks” ซึ่งเป็นการตั้งชื่อล้อกับบทความอันโด่งดังของ Eugene Wigner นักฟิสิกส์รางวัลโนเบล ที่รู้สึกฉงนใจเป็นอย่างมากว่าทำไมคณิตศาสตร์ถึงสามารถนำมาใช้กับวิทยาศาสตร์ได้ดียิ่งนัก และในบทความของ Karpathy ก็มีความน่าพิศวงปรากฏอยู่เช่นเดียวกัน ว่าทำไมเมื่อใช้ RNN มาเรียนรู้แบบจำลองภาษาระดับอักขระ (character-level language model) แล้วจึงสามารถสร้างลำดับของตัวอักษรขึ้นมาจนคล้ายกับภาษาของจริงได้ถึงเพียงนี้ ในบทความนี้ Karpathy ได้ทดลองใช้ RNN แบบ LSTM มาเรียนรู้แบบจำลองภาษากับ 5 ชุดข้อมูล อันได้แก่ Paul Graham Shakespeare Wikipedia Algebraic Geometry Linux Source Code ซึ่งหลังจากที่ได้เรียนรู้แล้ว แบบจำลองสามารถผลิตลำดับของตัวอักษรออกมาได้ละม้ายคล้ายจริงมาก ยกตัวอย่างเช่น Latex เกี่ยวกับ algebraic geometry ที่ RNN สร้างมาดังรูปด้านล่าง สำหรับผู้เขียนที่ไม่มีความรู้ทางด้านนี้ ดูเผินๆ แล้วนึกว่าเป็นของจริงเลยทีเดียว (คล้ายของจริงถึงขนาดมีคำว่า Proof. Omitted. อีกต่างหาก!) ตัวอย่างข้อความเกี่ยวกับ algebraic geometry ที่แบบจำลองสร้างขึ้น [ภาพจาก Karpathy] สำหรับชุดข้อมูล Wikipedia ชุดข้อมูล Algebraic Geometry และชุดข้อมูล Linux Source Code พบว่าข้อความที่แบบจำลองสร้างออกมามี syntax ที่เป๊ะมาก เช่นมีวงเล็บเปิดวงเล็บปิดและการเว้นวรรคเว้นบรรทัดที่ถูกต้อง มี tag ต่างๆ เหมือนจริง ส่วนชุดข้อมูล Paul Graham และ Shakespeare ถ้าดูเผินๆ ก็จะเหมือนกับผลงานที่สองท่านนี้ได้เขียนไว้เอง นอกจากนี้ Karpathy ยังได้วิเคราะห์ลึกถึงระดับเซลล์ของ LSTM ซึ่งเปรียบเสมือนหน่วยความจำที่ส่งผ่านข้อมูลเข้าประมวลผลในช่วงเวลาต่างๆ และพบว่าบางเซลล์เก็บข้อมูลที่มีความหมาย สามารถแปลผลได้โดยตรง เช่น มีเซลล์ที่บ่งบอกสถานะว่าตอนนี้อยู่ที่ตำแหน่งไหนของบรรทัด หรือบางเซลล์ก็บ่งชี้ว่ากำลังอยู่ในอัญประกาศ ตัวอย่างเช่นในรูปด้านล่าง ที่เป็นสีอ่อนคือเซลล์นี้มีค่าใกล้ๆ 0 และที่เป็นสีแก่แสดงว่าเซลล์นี้มีค่าสูง ซึ่งหมายความว่าเซลล์นี้กำลังตื่นตัวอยู่ ส่วนสีน้ำเงินกับสีแดงคือการแสดงค่าของเซลล์ที่เป็นบวกและลบครับ ซึ่งนับว่าเป็นการวิเคราะห์และการทำ visualization ที่เหนือชั้นมากเลย visualization ของเซลล์ใน LSTM [ภาพจาก Karpathy] จากผลลัพธ์ของการเรียนรู้แบบจำลองภาษาและผลการวิเคราะห์ข้างต้นนี้แสดงให้เห็นพลังความสามารถของ deep learning เป็นอย่างยิ่ง ซึ่ง Karpathy ได้บรรยายถึงสิ่งที่สวยงามเกี่ยวกับ LSTM ไว้ดังนี้ครับ Again, what is beautiful about this is that we didn’t have to hardcode at any point that if you’re trying to predict the next character it might, for example, be useful to keep track of whether or not you are currently inside or outside of quote. We just trained the LSTM on raw data and it decided that this is a useful quantity to keep track of. In other words, one of its cells gradually tuned itself during training to become a quote detection cell, since this helps it better perform the final task. This is one of the cleanest and most compelling examples of where the power in Deep Learning models (and more generally end-to-end training) is coming from. สุนทรภู่ บทประพันธ์ของ Shakespeare คือชุดข้อมูลชุดนึงที่ Karpathy ได้เลือกใช้มาสอนแบบจำลองภาษา ทำให้ผู้เขียนมีความสนใจอยากจะทดลองว่า ถ้าเป็นบทร้อยกรองของไทย อันมีฉันทลักษณ์ที่เคร่งครัดกว่า แบบจำลองภาษาจะยังสามารถเรียนรู้สิ่งเหล่านี้ได้หรือไม่ โดยในที่นี้จะขอเลือกรูปแบบร้อยกรองที่แพร่หลายและเป็นที่นิยมมากที่สุด นั่นคือกลอนสุภาพของสุนทรภู่มาครับ อย่างหม่อมฉันอันที่ดีและชั่ว ถึงลับตัวแต่ก็ชื่อเขาลือฉาว เป็นอาลักษณ์นักเลงทำเพลงยาว เขมรลาวลือเลื่องถึงเมืองนคร ถ้าในต่างประเทศมี Shakespeare ผู้เป็นเลิศด้านการใช้ภาษาอังกฤษ รังสรรค์บทละครและบทกวีที่ทรงคุณค่าไว้มากมาย ฝั่งเมืองไทยก็เห็นจะมีสุนทรภู่ ผู้แต่งกลอนโดยใช้ภาษาไทยได้อย่างวิจิตรแพรวพราว สามารถสถาปนากลอนแปดให้ยิ่งใหญ่ในโลกของคำประพันธ์และวรรณคดีไทยได้ตั้งแต่สมัยต้นกรุงรัตนโกสินทร์จวบจนกระทั่งถึงปัจจุบัน โดยฉันทลักษณ์ของกลอนสุนทรภู่จะเป็นดังนี้ครับ ฉันทลักษณ์ของกลอนแปด ในกลอนแปดบทนึงประกอบด้วย ๒ บาท บาทละ ๒ วรรค วรรคละ ๘ คำ (แต่อาจเป็น ๗ หรือ ๙ คำได้) สัมผัสนอกจะเป็นไปตามภาพ ส่วนสัมผัสในแม้จะไม่บังคับ แต่ถ้ามีด้วยก็จะทำให้บทกลอนของเราสละสลวยยิ่งขึ้น และในหัวข้อต่อๆ ไป เราจะ … มารู้จักอักขราอาร์เอ็นเอ็น โดยลองเล่นแต่งเป็นกลอนสุนทรภู่ แบบจำลองภาษาคงน่าดู ได้เรียนรู้เอไอไปพร้อมกัน แบบจำลองภาษาและอักขราอาร์เอ็นเอ็น แบบจำลองภาษา แบบจำลองภาษา (language model) ในที่นี้ คือ probabilistic model ที่รับเป็น sequence เข้ามา แล้วดูว่ามีความน่าจะเป็นที่ sequence นี้จะอยู่ในภาษาที่แบบจำลองเรียนรู้มามากน้อยแค่ไหน หรืออาจเขียนได้ว่าแบบจำลองภาษาจะให้ค่าของ ออกมา เมื่อ u คือหน่วยทางภาษาที่เราสนใจ ซึ่งโดยส่วนใหญ่อยู่ในระดับคำ แต่ที่จะใช้ในบทความนี้จะเป็นระดับอักขระครับ ถ้าเรามองว่า u เข้ามาเรียงกัน โดยตัวที่ 1 เข้ามาก่อน ต่อด้วยตัวที่ 2 ไปจนจบตัวที่ T จากค่า P ข้างต้น เมื่อใช้ product rule ของ probability จะได้ว่าค่านี้เท่ากับ ความน่าจะเป็นในการเกิด u ตัวที่ 1 คูณด้วยความน่าจะเป็นในการเกิด u ตัวที่ 2 เมื่อก่อนหน้านี้คือ u ตัวที่ 1 แล้วคูณไปเรื่อยๆ ด้วยความน่าจะเป็นในการเกิด u ตัวต่อไป เมื่อเกิด u ตัวก่อนๆ หน้าขึ้นมาแล้ว ซึ่งจะเขียนเป็นสมการได้ว่า แทบทุกงานทางด้าน machine learning ที่มีภาษาเข้ามาเกี่ยวข้อง แบบจำลองภาษาจะเข้ามาช่วยได้ ไม่ว่าจะเป็นงานเช่น machine translation เรื่อยไปจนถึง speech recognition และ OCR โดยในสมัยก่อนงานเหล่านี้จะมีหลาย module ประกอบกัน และส่วนใหญ่แบบจำลองภาษาจะถูกปลั๊กเข้าไปข้างหลัง เป็น module ตัวสุดท้ายเพื่อแก้ไขผลลัพธ์ที่ออกมาให้มีความถูกต้องมากยิ่งขึ้น แต่ในปัจจุบันงานเหล่านี้จะใช้ deep learning ที่เป็น end-to-end ซึ่งแบบจำลองภาษาจะถูกเรียนรู้ขึ้นมาเสร็จสรรพโดยอัตโนมัติใน neural network ขนาดใหญ่นี้เลย และนอกจากจะนำไปใช้ช่วยเหลือในงานอื่นแล้ว แบบจำลองภาษาด้วยตัวของมันเองก็อาจเอามาใช้ผลิต sequence ออกมาเล่นๆ เหมือนอย่างในบทความนี้ก็ได้ครับ Statistical Language Model การสร้างแบบจำลองภาษาแบบดั้งเดิม จะใช้วิธีการทางสถิติ นั่นคือค่า P ของ sequence ต่างๆ จะคำนวณมาจากการนับคำที่ปรากฏอยู่ใน corpus โดยตรง ค่า P ข้างต้นมักประมาณโดยใช้ข้อมูลแค่ n ตัว หรือที่เรียกว่า n-gram ดังนี้ ถึงแม้ n จะมีค่าไม่มากแล้ว แต่ sequence ของ n-gram ที่มีโอกาสเกิดขึ้นได้ในภาษาจริงนั้นก็มีมากมายเกินกว่าที่จะอยู่ใน corpus ใดๆ การให้ค่า count เป็น 0 สนิทเลยจึงไม่ใช่เรื่องที่ดี ทำให้ต้องมีเทคนิคเช่น smoothing หรือ back-off เข้ามาช่วยด้วยครับ Neural Language Model ในปี ค.ศ. 2003 Bengio et al. ได้นำ feedforward neural network มาใช้สร้างแบบจำลองภาษา เปเปอร์นี้มีเรื่อง word embedding มาก่อน word2vec และมาก่อนการใช้ RNN สำหรับ neural language model นับสิบปี ต่อมา Sutskever et al. และ Graves ได้ใช้ RNN และ LSTM สำหรับสร้างแบบจำลองภาษา ซึ่งเป็นรากฐานก่อนที่จะมาถึง Karpathy นี่เองครับ ใน neural language model นี้ ค่า P จะออกมาจาก softmax นั่นคือ เมื่อ V เป็นเซ็ตของหน่วยทางภาษา เช่นคำทั้งหมดที่ใช้ ส่วน h จะเป็นค่าที่ออกจาก hidden state ชั้นสุดท้าย และ v คือ output embedding ที่แม็พจาก h มาสู่ output ซึ่งเป็นหน่วยทางภาษาที่เราสนใจ neural language model จะได้เปรียบตรงที่ information ทั้งหลายถูกเก็บอยู่ใน hidden state ซึ่งเป็น distributed representation ขณะที่ statistical language model จะทำการนับ discrete term ต่างๆ โดยตรง จึงเจอปัญหาการระเบิดแบบ exponential เพียงแค่เพิ่มค่าของ n จำนวนพจน์ต่างๆ ที่ต้องคำนวณจะมีเพิ่มขึ้นอย่างมหาศาล ทำให้เสมือนว่า neural language model สามารถใช้ information ย้อนไปได้ไกลกว่า การวัดผล ค่ามาตรฐานสำหรับการวัดผลของแบบจำลองภาษาคือ cross entropy ซึ่งเป็นการวัดความเคลื่อนคลาด ระหว่าง probability distribution ที่แบบจำลองทำนาย (P) กับสิ่งที่เกิดขึ้นจริง (Q) และเนื่องจากสิ่งที่เกิดขึ้นจริงคือหน่วยทางภาษาหน่วยเดียว เช่นเป็นคำๆ นึง จากคำที่เป็นไปได้ทั้งหมด ค่า Q ของคำๆ นั้นจึงเป็น 1 และค่าที่เหลือเป็น 0 หมด ทำให้ลดรูปได้เป็นบรรทัดสุดท้าย ในสมการด้านล่างนี้ สำหรับแบบจำลองระดับอักขระ ในตอนทดสอบจะเรียกค่า cross entropy นี้ว่า bit per character (bpc) ส่วนในตอนเรียนรู้ ค่านี้ก็คือ loss ที่ใช้กับ softmax layer ครับ อักขราอาร์เอ็นเอ็น (คำนี้เป็นคำที่ผู้เขียนประดิษฐ์ขึ้นเองเพื่อใช้ในบทความนี้นะครับ โดยจะหมายถึง RNN ที่ใช้สำหรับสร้าง character-level language model เนื่องจากคำว่า character มีศัพท์บัญญัติภาษาไทยว่า อักขระ จึงจับมาสนธิกับคำว่า อาร์เอ็นเอ็น ที่เป็นคำทัพศัพท์เสียเลย) ตัวอย่างอย่างง่าย ตัวอย่างอย่างง่ายเพื่ออธิบายการทำงานของอักขราอาร์เอ็นเอ็นจะเป็นดังรูปนี้ครับ ตัวอย่างการทำงานของอักขราอาร์เอ็นเอ็น [ภาพจาก Karpathy] ในตัวอย่างนี้จะมีตัวอักษรแค่ 4 ตัว คือ ‘h’ ‘e’ ‘l’ และ ‘o’ ด้าน input vector และ output vector จึงมี 4 มิติ โดยที่ค่าในแต่ละมิติจะสัมพันธ์กับตัวอักษรแต่ละตัว ซึ่ง input vector ในที่นี้จะเป็น one-hot encoding ส่วน output vector ในแต่ละตำแหน่งจะแสดงถึงความมั่นใจในการตอบ ‘h’ ‘e’ ‘l’ และ ‘o’ ออกมา ซึ่งในการเรียนรู้ จะต้องทำให้ค่าในตำแหน่งสีเขียวมีค่ามาก และค่าในตำแหน่งสีแดงมีค่าน้อย เพื่อให้คำตอบออกมาถูกต้องตามข้อมูลที่สอนเข้าไป RNN จะทำงานทีละ time step โดยรับ input เป็นตัวอักษรที่เวลานั้นเข้ามา แล้ว hidden state จะคำนวณค่าจาก input vector ที่เข้ามาเวลานั้นและจาก hidden state ก่อนหน้า จากนั้นจะให้ output เป็นการทำนายตัวอักษรตัวถัดไป หรือเขียนเป็นสมการได้ว่า ของที่ใช้จริง ถ้านำ RNN ข้างต้นไปใช้จริงจะพบว่าทำการเรียนรู้ได้ไม่ดี เนื่องจากมีปรากฏการณ์ vanishing/exploding gradient เป็นปัญหาใหญ่อยู่ ในการทดลองนี้จึงได้ใช้ RNN แบบ LSTM และการตั้งค่าเกือบทั้งหมดจะเหมือนกับที่ Karpathy ทำ แต่ก็ไม่ได้นำโค้ดของ Karpathy มาใช้โดยตรงครับ เนื่องจากผู้เขียนไม่รู้ภาษา Lua จึงได้ใช้โค้ด Keras ของคุณ Yu Xuan Tay มาแทน โดยทำการเพิ่มเติมแก้ไขในจุดสำคัญๆ สองส่วน คือทำให้สามารถรับภาษาไทยได้ กับเพิ่มในส่วนของการทำ validation และ testing สำหรับ RNN architecture ที่ใช้ในการทดลอง สามารถอธิบายได้ดังนี้ อักขราอาร์เอ็นเอ็นที่ใช้ในการทดลอง input ที่เข้ามาจะผ่าน embedding layer ก่อน เพื่อแปลงจาก one-hot encoding ตรงๆ ให้อยู่ใน embedding space ที่สื่อความหมายได้ดีขึ้น กล่องของ recurrent layer จะเหมือนกับ hidden layer ในรูปก่อนหน้า เพียงแต่เปลี่ยนจาก RNN ธรรมดาเป็น LSTM และมีหลายชั้น ใน Keras เมื่อระบุให้ return_sequences=True ที่ชั้น LSTM ในแต่ละเวลาจะคาย output ออกมาด้วย ถ้ามี LSTM ซ้อนกันหลายชั้น ก็จะใช้ output ของ LSTM ชั้นล่าง มาเป็น input ของ LSTM ที่ชั้นถัดไป ในเวลาเดียวกัน ใน Keras จะมี TimeDistributed layer สำหรับครอบ output ที่ออกมาจาก LSTM ชั้นสุดท้ายในแต่ละเวลา โดยในที่นี้จะครอบเป็น fully connected layer หรือเรียกอีกชื่อว่า dense layer ที่มี output ขนาดเท่ากับจำนวนตัวอักษรที่ใช้ และให้ softmax เป็น activation function ใน Keras ถ้า stateful=True จะให้ค่าของ hidden state มีความต่อเนื่องไประหว่าง minibatch ด้วย ซึ่งจำเป็นต้องใช้ในงานนี้เพราะว่าข้อมูลที่ใช้สอนแบบจำลองภาษามีความต่อเนื่องกันไปโดยตลอด ค่าของ hidden state จึงควรจะต่อเนื่องตามไปด้วย มิใช่เริ่มจาก 0 ใหม่ในทุกๆ minibatch และเมื่อเซ็ตให้ใช้ stateful แล้ว กระบวนการเตรียมข้อมูลก่อนนำเข้าไปเทรนจะซับซ้อนกว่าแบบทั่วไปที่เป็น stateless อยู่เล็กน้อยครับ ในส่วนของ source code ที่ใช้ ได้เปิดเป็น open source ไว้ที่นี่ครับ การทดลองสร้างแบบจำลองภาษา ชุดข้อมูล ชุดข้อมูลภาษาไทยในที่นี้มาจากวรรณคดีเรื่องพระอภัยมณี โดยผู้เขียนได้ใช้เฟรมเวิร์ค Scrapy ดึงข้อมูลมาจากห้องสมุดดิจิทัลวชิรญาณ ซึ่งต้องขอขอบคุณทางห้องสมุดมาไว้ ณ ที่นี้ครับ เพราะการจัดทำวรรณกรรมของไทยให้อยู่ในรูปแบบหนังสืออิเล็กทรอนิกส์นอกจากจะมีคุณูปการอย่างมหาศาลในด้านการอนุรักษ์แล้ว ยังมีประโยชน์ต่อวงการ NLP ภาษาไทยอีกด้วย ข้อมูลชุดนี้จะใช้ชื่อว่า phra_aphai สิริแล้วมี 2,612,950 ตัวอักษร บันทึกไฟล์โดยใช้ encoding เป็น UTF-8 โดยบรรทัดนึงจะมีสองวรรค แต่ละวรรคคั่นด้วยตัวอักษรช่องว่าง ตัวอย่างบทกลอนที่อยู่ในข้อมูลฝึกสอนเป็นดังนี้ครับ ชุดข้อมูล phra_aphai มีขนาดประมาณ 2M การทดลองนี้จึงได้นำชุดข้อมูล tinyshakespeare ที่มีขนาดประมาณ 1M และชุดข้อมูล shakespeare ที่มีประมาณ 4M มาใช้ทดลองเพื่อทำการเปรียบเทียบกันด้วย โดยมีการแบ่งข้อมูลสำหรับฝึกสอนและทดสอบดังตารางด้านล่างนี้ครับ โดยชุดข้อมูล tinyshakespeare และ shakespeare จะตัดตัวอักษรตำแหน่งที่ 1M และ 4M จนถึงตัวสุดท้ายไปเป็น test data ตามลำดับ และที่เหลือจะแบ่งข้อมูล 10% ข้างท้ายมาเป็น validation data ส่วนนอกนั้นจะเป็น training data สำหรับชุดข้อมูล phra_aphai จะตัดตอนที่ ๑๑๖ ไปจนจบ เป็น test data และตอนที่ ๑๐๔ ถึง ๑๑๕ เป็น validation data เซ็ตอัพ ในการทดลองมีการตั้งค่า hyperparameter ต่างๆ เป็นดังนี้ network parameters num_layer = 3 rnn_size = 512 ในที่นี้จะใช้ LSTM ซ้อนกันสามชั้น และ hidden state เป็น vector ขนาด 512 จากการทดลองพบว่าค่า rnn_size มีความสำคัญมากทีเดียว ในช่วงแรกผู้เขียนเซ็ตเป็นขนาด 128 แล้วผลลัพธ์ที่ออกมาไม่ค่อยดีเท่าไหร่ จึงได้รู้ว่าเป็นค่าที่น้อยไป input parameters batch_size = 64 seq_len = 256 embedding_size = 32 ผู้เขียนสันนิษฐานว่าเมื่อเป็น stateful แล้ว ค่า seq_len อาจไม่สำคัญเท่าไหร่ แต่ก็เลือกใช้ seq_len เป็น 256 ซึ่งจากการสุ่มตัวอย่างดูพบว่าค่านี้ครอบคลุมกลอนสองบทได้ สำหรับ parameter อื่น จะให้ minibatch นึงมีข้อมูล 64 ท่อน และขนาด vector ที่ออกมาจาก embedding layer เท่ากับ 32 optimization Adam optimizer lr = 0.001 beta_1 = 0.9 beta_2 = 0.999 decay = 0 ในที่นี้จะทำ optimization โดยใช้วิธี Adam optimizer ส่วน parameter อื่นๆ ก็ใช้ค่าที่เป็น default ของการทำ optimization วิธีนี้ครับ regularization drop_rate = 0.5 clip_norm = 5 ในที่นี้จะใช้ค่า dropout rate เท่ากับ 50% ซึ่งเป็นค่ามาตรฐานที่ส่วนใหญ่นิยมใช้กัน และเพื่อป้องกันปัญหา exploding gradient ของ RNN จะมีการขริบ gradient ไม่ให้มีขนาดที่เกิน 5 สรุป โครงสร้างของ neural network และจำนวน parameter ทั้งหมดสามารถสรุปได้ดังตารางนี้ Karpathy ได้ให้เคล็ดลับในการสร้าง neural network ไว้นิดนึงว่าจำนวน parameter ที่มีทั้งหมด กับจำนวนข้อมูลควรจะอยู่ใน order เดียวกัน ในที่นี้จะเห็นว่าทั้ง parameter และชุดข้อมูลของเราอยู่ที่สเกลระดับ 10⁶ เหมือนกัน จึงน่าจะโอเคระดับนึง ปัญหา underfitting คงจะไม่มี แต่อาจต้องคอยจับตาปัญหา overfitting อยู่บ้างครับ ผลการทดลอง ในที่นี้จะนำเสนอผลการทดลองทั้งแบบที่เป็นตัวเลข (quantity) และตัวอักษร (quality) นะครับ ตัวเลข ในการทดลองนี้ได้ทำการเทรนทั้งหมด 1000 รอบ โดยใช้ Tesla K80 GPU แล้วทดสอบกับ validation data ทุกๆ 10 รอบ เมื่อเทรนเสร็จแล้วจะนำแบบจำลองของรอบที่ให้ค่า bpc ใน validation data ต่ำสุด มาใช้ทดสอบกับ test data ซึ่งให้ผลดังตาราง พบว่าค่า bpc ของ test data ขึ้นอยู่กับขนาดของข้อมูลโดยตรง โดยชุดข้อมูลที่มีข้อมูลมากกว่า จะให้ค่า bpc ที่ดีกว่า สำหรับค่า bpc ของชุดข้อมูล phra_aphai ที่ประมาณ 2 ในเชิง information theory อาจตีความได้ว่า ถ้าเราต้องการจะบีบอัดข้อมูลหนังสือพระอภัยมณีทั้งเล่ม ในทางปฏิบัติอาจสามารถทำได้เพียงใช้แค่ 2 บิต ต่อจำนวนตัวอักษรหนึ่งตัว ก็จะลดขนาดไฟล์หนังสือพระอภัยมณีลงได้ราวๆ 4 เท่า เมื่อเทียบกับการ encoding ทั่วไปที่ใช้ 8 บิต ค่า bpc นี้จะเป็น upper bound ของ optimal code จริงๆ ซึ่ง optimal code นี้ก็คือ lower bound ของ data compression ที่เป็นไปได้นั่นเองครับ ส่วน learning curve ของชุดข้อมูลต่างๆ จะเป็นดังรูปด้านล่างนี้ learning curve ของสามชุดข้อมูล และเมื่อเอาเฉพาะ validation data ของสามชุดข้อมูลมาเทียบกัน จะเป็นดังนี้ ค่า bpc ของ validation data จะเห็นว่าชุดข้อมูล phra_aphai ลู่เข้าช้ากว่าชุดข้อมูล shakespeare ทั้งสองอยู่เล็กน้อย โดยค่า bpc ที่ดีที่สุดใน validation data มาช้ากว่าอยู่ประมาณร้อยรอบ เมื่อดูกราฟแล้วพบว่าค่า bpc ของ validation data มีอาการหวั่นไหวในรอบหลังๆ โดยเฉพาะชุดข้อมูล tinyshakespeare ที่สั่นค่อนข้างแรง อาจจะเป็นเพราะผู้เขียนไม่ได้ปรับ learning rate ให้มี decay และจากการที่ค่า bpc ของ validation data กับ test data ในชุดข้อมูล tinyshakespeare มีค่าค่อนข้างต่างกัน ทำให้เห็นว่าชุดข้อมูลนี้มีข้อมูลจำนวนน้อยเกินไป และเกิดปัญหา overfitting อาจจะต้องปรับเรื่อง regularization สำหรับชุดข้อมูลนี้ให้มีมากขึ้น สำหรับผลของการเปลี่ยน hyperparameter ต่างๆ จะมีบล็อกนี้ที่ได้ทำการทดลองไว้ครับ ตัวอักษร สุดท้ายนี้เราจะให้ RNN ที่เรียนรู้แบบจำลองภาษามาแล้ว สร้างลำดับของตัวอักษรต่อๆ กันไปเรื่อยๆ โดยตัวอักษรที่ถูกสร้างขึ้นในเวลาก่อนหน้า จะถูกนำไปใช้เป็น input สำหรับสร้างตัวอักษรในเวลาถัดไป และจะสุ่มตัวอักษรจาก 3 ตัวที่มีค่า probability สูงสุด เป็น output ออกมา โดยใช้ seed string หรือ string เริ่มต้นเป็น “๏ ก” กลอนที่ได้ 5 บทแรกเป็นดังนี้ครับ ในด้านฉันทลักษณ์ส่วนใหญ่ทำได้ค่อนข้างดี ถ้ามองเพียงผิวเผินอาจนึกว่าสุนทรภู่มาเอง แบบจำลองนี้สามารถเรียนรู้ข้อจำกัดด้านจำนวนพยางค์ การเว้นวรรค และการเว้นบรรทัด แล้วแต่งกลอนออกมาตามข้อกำหนดนี้ได้เป๊ะมาก สัมผัสในไม่ด่างพร้อย ถ้าดูแค่แต่ละวรรคจะเห็นว่าค่อนข้างสมบูรณ์ในตัว แต่สัมผัสนอกบางครั้งยังไม่ค่อยจะได้ มีศัพท์ใหม่บางคำที่แบบจำลองสร้างขึ้นมาเอง เช่นคำว่า “สำอาญ” ซึ่งดูจะคล้ายๆ คำว่า สำราญ ผสมกับ สำอาง (วรรคที่ว่า “แม้นมารดรอดอ่านสำอาญตา” ผู้เขียนอ่านครั้งแรกแล้วรู้สึกว่าตัดคำได้ยากมากครับ) วรรคแรกสุดในหนังสือพระอภัยมณีเป็นวรรครับ (วรรคที่ ๒ ในกลอน ๑ บท) ซึ่งแบบจำลองก็เรียนรู้ได้ และแต่งออกมาแบบนี้เหมือนกัน เมื่อขึ้นต้นบทกลอนหัวข้อใหม่ จะใช้เครื่องหมายตาไก่หรือฟองมัน (๏) และจบท้ายหัวข้อนั้นด้วยไปยาลน้อย (ฯ) ซึ่งแบบจำลองก็ใช้สองเครื่องหมายนี้มาต่อกันได้ไม่พลาดเลย ตัวอย่างเช่นกลอนอีกช่วงนึงที่แบบจำลองแต่งออกมาดังนี้ จะเห็นว่าในส่วนสัมผัสนอกของบาทเดียวกัน แบบจำลองนี้พยายามจะทำให้ได้อยู่ เช่นในบทที่ ๔ ของกลอนด้านบน มีการเลือกใช้คำว่า “ตับ” เพื่อให้สัมผัสกับคำว่า “รับ” และพยายามเติม ม.ม้า ให้เป็นคำว่า “ว่าม” เพื่อให้สัมผัสกับคำว่า “งาม” แต่สัมผัสนอกที่อยู่คนละบาทนั้นยังทำได้ไม่ค่อยดี ส่วนในด้านความหมาย หรือการจะให้แต่งอะไรออกมาเป็นเรื่องเป็นราวนั้น แน่นอนว่ายังเกินความสามารถของแบบจำลองภาษาลักษณะนี้อยู่ครับ การที่แบบจำลองยังทำสัมผัสนอกและสืบทอดความหมายระหว่างวรรคได้ไม่ดี แสดงให้เห็นว่าเรื่อง long-range dependency ยังเป็นปัญหาที่สำคัญ และรอคอยการแก้ไขต่อไป ท่านผู้อ่านสามารถดูข้อมูลที่แบบจำลองผลิตมาจำนวนหนึ่งล้านตัวอักษรได้ที่นี่ และลองเทียบกับวรรณคดีเรื่องพระอภัยมณีของจริงได้ที่นี่ครับ มีทฤษฎีลิงอนันต์ได้กล่าวไว้ว่า ถ้าเราจับลิงมาตัวนึง ให้นั่งเคาะแป้นพิมพ์ไปเรื่อยๆ ซักวันก็คงได้บทประพันธ์ของ Shakespeare ออกมา แต่คงต้องรอนาน (มากๆ) แต่ถ้าหากเราให้ deep learning แทนลิงตัวนั้น ก็จะสามารถแต่งอะไรที่คล้ายกับ Shakespeare หรือสุนทรภู่ออกมาได้ โดยที่ไม่ต้องใช้เวลาถึงอสงไขยอย่างในทฤษฎีลิงอนันต์ครับ ปิดท้าย Karpathy ถ้าคนธรรมดาจะแต่งกลอนซักบทให้ได้สละสลวยอาจต้องใช้เวลาเป็นวัน ว่ากันว่าสุนทรภู่ช่วงพีคพูดกลอนออกมาแล้วใช้เสมียนสองคนนั่งจดยังจดแทบไม่ทัน แต่ถึงกระนั้น วรรณคดีเรื่องพระอภัยมณีกว่าจะแต่งเสร็จก็ใช้เวลามากกว่า 20 ปี ในปัจจุบันนี้ก็ยังมีปัญหาว่านักอ่านรอเสพผลงานของนักเขียนที่ตนเองชื่นชอบอย่างกระวนกระวาย หลายคนถึงกับภาวนาว่าขออย่าให้ตนเองและนักเขียนตายไปก่อนที่นิยายหรือมังงะที่ติดตามอยู่จะถึงตอนจบเลย เพราะฉะนั้น จะดีแค่ไหนถ้าความฝันใฝ่ของ Karpathy เป็นจริง จะดีแค่ไหนถ้า AI สามารถแต่งนิยายชุด A Song of Ice and Fire เล่มใหม่ หรือวาดมังงะ Hunter x Hunter ตอนใหม่ได้เพียงในเวลาไม่กี่วัน แทนที่จะต้องรอคอยกันนานหลายๆ ปี อย่างที่เป็นอยู่ Karpathy มีความเชื่อว่าซักวันนึง แม้แต่หนังทั้งเรื่องก็สามารถใช้ AI สร้างขึ้นมาได้ ใจนึงผู้เขียนคิดว่าถ้า extrapolation จากความก้าวหน้าแบบก้าวกระโดดของโลก AI ในปัจจุบัน เวลานั้นคงจะมาถึงในอีกไม่ช้า แต่อีกใจนึงผู้เขียนก็คิดว่ายังมีช่องว่างอันกว้างใหญ่ที่จำเป็นต้องข้ามไปอยู่ ผู้เขียนเชื่อว่าในสมองของคนเรานี้มีแบบจำลองภาษาอยู่อย่างแน่นอน แต่เรื่องความคิดสร้างสรรค์นั้นเป็นอะไรที่มากกว่าแบบจำลองภาษา กว่าจะกลั่นถ้อยคำออกมาเป็นหนังสือซักเล่ม บทความซักบท หรือเรื่องราวซักเรื่อง มนุษย์เราใช่ว่าสักแต่จะนำตัวอักษรหรือคำที่น่าจะเป็นมาเรียงร้อยต่อกัน แต่ยังมีเรื่องของการวางแผน การใช้เหตุผล การใช้ภาษา อารมณ์ศิลปะ รวมทั้งจะต้องประมวลความรู้ทางโลก ความรู้สึกนึกคิด และประสบการณ์ชีวิตที่มีทั้งหมดเข้าด้วยกัน ซึ่งสิ่งเหล่านี้ยังเกินกว่าความสามารถของ AI ในปัจจุบัน แม้แต่ในงานบางอย่างที่ AI ทำได้ดีแล้ว เราก็ยังไม่เข้าใจกลไกข้างในนัก อย่างที่ Karpathy ใช้คำว่า unreasonable effectiveness นั่นเองครับ สุนทรภู่ ย้อนกลับไปเมื่อสมัยต้นกรุงรัตนโกสินทร์ เราจะได้เห็นภาพชีวิตกวีเอกของไทยที่ได้ดีและตกอับสลับกัน แต่ไม่ว่าจะเป็นช่วงไหน สุนทรภู่ก็ยังคงเขียนกลอนอยู่อย่างต่อเนื่อง เฉพาะเรื่องพระอภัยมณีก็นับว่ามีความยาวสูงเพียงพอที่จะให้ผู้เขียนนำมาใช้ทดลองสร้าง language model เทียบกับงานของ Shakespeare ได้ เนื่องในโอกาสวันสุนทรภู่ปีนี้ ผู้เขียนขอน้อมรำลึกถึงท่านด้วยครับ อนุสาวรีย์สุนทรภู่ [ภาพจาก Wikipedia] หนึ่งขอฝากปากคำทำหนังสือ ให้สืบชื่อชั่วฟ้าสุธาสถาน สุนทราอาลักษณ์เจ้าจักรพาฬ พระทรงสารศรีเศวตเกศกุญชร หมายเหตุ บทความนี้จะกลายเป็นส่วนหนึ่งของเว็บไซต์ ThAIKeras (มาเรียนรู้ Keras และ AI ไปด้วยกัน) กับ mena.ai (แพลตฟอร์มสำหรับสร้าง chatbot ภาษาไทย) ทางทีมงาน ThAIKeras และ mena.ai จะทยอยเขียนบทความเพื่อเสริมสร้างความรู้ความเข้าใจในด้าน AI ต่อไปครับ

มารู้จักอักขราอาร์เอ็นเอ็น โดยลองเล่นแต่งเป็นกลอนอักษรสาร

32

มารู้จักอักขราอาร์เอ็นเอ็น-โดยลองเล่นแต่งเป็นกลอนอักษรสาร-1264028567a5

2018-06-30

2018-06-30 03:41:43

https://medium.com/s/story/มารู้จักอักขราอาร์เอ็นเอ็น-โดยลองเล่นแต่งเป็นกลอนอักษรสาร-1264028567a5

false

1,421

null

Deep Learning

deep-learning

Deep Learning

12,189

ppp

null

a4b87a47dc5c

u41ppp

15

1

20,181,104

null