paper_id
stringlengths 10
10
⌀ | yes_no
bool 2
classes | paper_index
int64 0
519
⌀ | evidence
stringlengths 0
37.7k
| question
stringlengths 4
11.7k
| answer
stringlengths 1
26k
|
|---|---|---|---|---|---|
null | false
| 292
|
Neural machine translation (NMT) BIBREF0 is a promising paradigm for extracting translation knowledge from parallel text. NMT systems have achieved competitive accuracy rates under large-data training conditions for language pairs such as English-French. However, neural methods are data-hungry and learn poorly from low-count events. This behavior makes vanilla NMT a poor choice for low-resource languages, where parallel data is scarce. Table 1 shows that for 4 low-resource languages, a standard string-to-tree statistical MT system (SBMT) BIBREF1 , BIBREF2 strongly outperforms NMT, even when NMT uses the state-of-the-art local attention plus feed-input techniques from luong2015effective.
In this paper, we give a method for substantially improving NMT results on these languages. Neural models learn representations of their input that are often useful across tasks. Our key idea is to first train a high-resource language pair, then use the resulting trained network (the parent model) to initialize and constrain training for our low-resource language pair (the child model). We find that we can optimize our results by fixing certain parameters of the parent model, letting the rest be fine-tuned by the child model. We report NMT improvements from transfer learning of 5.6 Bleu on average, and we provide an analysis of why the method works. The final NMT system approaches strong SBMT baselines in in all four language pairs, and exceeds SBMT performance in one of them. Furthermore, we show that NMT is an exceptional re-scorer of `traditional' MT output; even NMT that on its own is worse than SBMT is consistently able to improve upon SBMT system output when incorporated as a re-scoring model.
We start by a brief description of our NMT model in Section "NMT Background" . Section "Transfer Results" gives some background on transfer learning and explains how we use it to improve machine translation performance. Our main experiments translating Hausa, Turkish, Uzbek and Urdu into English with the help of a French-English parent model are presented in Section "Experiments" . Section "Analysis" explores alternatives to our model to enhance understanding. We find that the choice of parent language pair affects performance, and provide an empirical upper bound on transfer performance using an artificial language. We experiment with English-only language models, copy models, and word-sorting models to show that what we transfer goes beyond monolingual information and using a translation model trained on bilingual corpora as a parent is essential. We show the effects of freezing, fine-tuning, and smarter initialization of different components of the attention-based NMT system during transfer. We compare the learning curves of transfer and no-transfer models showing that transfer solves an overfitting problem, not a search problem. We summarize our contributions in Section "Conclusion" .
However, neural methods are data-hungry and learn poorly from low-count events. This behavior makes vanilla NMT a poor choice for low-resource languages,where parallel data is scarce.
|
We know that neural methods are data-hungry and learn poorly from low-count events, what limitations does this behavior have?
|
This behavior makes vanilla NMT a poor choice for low-resource languages, where parallel data is scarce.
|
null | false
| null |
The origins of the travelling salesman problem are unclear. A handbook for travelling salesmen from 1832 mentions the problem and includes example tours through Germany and Switzerland, but contains no mathematical treatment.
William Rowan Hamilton
The TSP was mathematically formulated in the 19th century by the Irish mathematician William Rowan Hamilton and by the British mathematician Thomas Kirkman. Hamilton's icosian game was a recreational puzzle based on finding a Hamiltonian cycle. The general form of the TSP appears to have been first studied by mathematicians during the 1930s in Vienna and at Harvard, notably by Karl Menger, who defines the problem, considers the obvious brute-force algorithm, and observes the non-optimality of the nearest neighbour heuristic:
We denote by messenger problem (since in practice this question should be solved by each postman, anyway also by many travelers) the task to find, for finitely many points whose pairwise distances are known, the shortest route connecting the points. Of course, this problem is solvable by finitely many trials. Rules which would push the number of trials below the number of permutations of the given points, are not known. The rule that one first should go from the starting point to the closest point, then to the point closest to this, etc., in general does not yield the shortest route.
It was first considered mathematically in the 1930s by Merrill M. Flood who was looking to solve a school bus routing problem. Hassler Whitney at Princeton University generated interest in the problem, which he called the "48 states problem". The earliest publication using the phrase "travelling salesman problem" was the 1949 RAND Corporation report by Julia Robinson, "On the Hamiltonian game (a traveling salesman problem)."
In the 1950s and 1960s, the problem became increasingly popular in scientific circles in Europe and the United States after the RAND Corporation in Santa Monica offered prizes for steps in solving the problem. Notable contributions were made by George Dantzig, Delbert Ray Fulkerson and Selmer M. Johnson from the RAND Corporation, who expressed the problem as an integer linear program and developed the cutting plane method for its solution. They wrote what is considered the seminal paper on the subject in which with these new methods they solved an instance with 49 cities to optimality by constructing a tour and proving that no other tour could be shorter. Dantzig, Fulkerson and Johnson, however, speculated that given a near optimal solution we may be able to find optimality or prove optimality by adding a small number of extra inequalities (cuts). They used this idea to solve their initial 49 city problem using a string model. They found they only needed 26 cuts to come to a solution for their 49 city problem. While this paper did not give an algorithmic approach to TSP problems, the ideas that lay within it were indispensable to later creating exact solution methods for the TSP, though it would take 15 years to find an algorithmic approach in creating these cuts. As well as cutting plane methods, Dantzig, Fulkerson and Johnson used branch and bound algorithms perhaps for the first time.
In 1959, Jillian Beardwood, J.H. Halton and John Hammersley published an article entitled "The Shortest Path Through Many Points" in the journal of the Cambridge Philosophical Society. The Beardwood–Halton–Hammersley theorem provides a practical solution to the travelling salesman problem. The authors derived an asymptotic formula to determine the length of the shortest route for a salesman who starts at a home or office and visits a fixed number of locations before returning to the start.
In the following decades, the problem was studied by many researchers from mathematics, computer science, chemistry, physics, and other sciences. In the 1960s, however, a new approach was created, that instead of seeking optimal solutions would produce a solution whose length is provably bounded by a multiple of the optimal length, and in doing so would create lower bounds for the problem; these lower bounds would then be used with branch and bound approaches. One method of doing this was to create a minimum spanning tree of the graph and then double all its edges, which produces the bound that the length of an optimal tour is at most twice the weight of a minimum spanning tree.
In 1976, Christofides and Serdyukov independently of each other made a big advance in this direction: the Christofides-Serdyukov algorithm yields a solution that, in the worst case, is at most 1.5 times longer than the optimal solution. As the algorithm was simple and quick, many hoped it would give way to a near optimal solution method. However, this hope for improvement did not immediately materialize, and Christofides-Serdyukov remained the method with the best worst-case scenario until 2011, when a (very) slightly improved approximation algorithm was developed for the subset of "graphical" TSPs. In 2020 this tiny improvement was extended to the full (metric) TSP.
Richard M. Karp showed in 1972 that the Hamiltonian cycle problem was NP-complete, which implies the NP-hardness of TSP. This supplied a mathematical explanation for the apparent computational difficulty of finding optimal tours.
Great progress was made in the late 1970s and 1980, when Grötschel, Padberg, Rinaldi and others managed to exactly solve instances with up to 2,392 cities, using cutting planes and branch and bound.
In the 1990s, Applegate, Bixby, Chvátal, and Cook developed the program Concorde that has been used in many recent record solutions. Gerhard Reinelt published the TSPLIB in 1991, a collection of benchmark instances of varying difficulty, which has been used by many research groups for comparing results. In 2006, Cook and others computed an optimal tour through an 85,900-city instance given by a microchip layout problem, currently the largest solved TSPLIB instance. For many other instances with millions of cities, solutions can be found that are guaranteed to be within 2–3% of an optimal tour.
|
What is the history of the travelling salesman problem throughout the years in a short bulleted list?
|
* 19th century - mathematically formulated for first time by the Irish mathematician William Rowan Hamilton and by the British mathematician Thomas Kirkman
* 1930s - First studied by mathematicians in Vienna and at Harvard, most notably by Karl Menger and Merill M. Flood who was looking to solve a school bus routing problem.
* 1959 - Jillian Beardwood, J.H. Halton and John Hammersley published an article entitled "The Shortest Path Through Many Points" in the journal of the Cambridge Philosophical Society, which provided the first practical solution.
* 1960s - new approach is created, which produce a solution whose length is provably bounded by a multiple of the optimal length.
* 1972 - Richard M. Karp showed that the Hamiltonian cycle problem was NP-complete, which implies the NP-hardness of traveling salesman problem.
* 1976 - Christofides and Serdyukov independently invented new algorithm, which yields a solution that in the worst case is at most 1.5 times longer than the optimal solution.
* 1990s - Applegate, Bixby, Chvátal, and Cook developed the program Concorde that has been used in many recent record solutions.
* 1991 - Gerhard Reinelt published the TSPLIB, a collection of benchmark instances of varying difficulty, which has been used by many research groups for comparing results
* 2006 - Cook and others computed an optimal tour through an 85,900-city instance given by a microchip layout problem, currently the largest solved TSPLIB instance.
* 2011 - Slightly improved approximation algorithm is developed for the subset of "graphical" traveling salesman problem.
* 2020 - The slightly improved approximation algorithm developed for the subset of "graphical" is extended to cover the full traveling salesman problem.
|
null | false
| null |
Categorize these bands as either 'rock', 'pop', 'metal' or 'indie'. Metallica, Queen, One direction, Korn, Coldplay, Arcade fire, backstreet boys
|
metal: Metallica, Korn
rock: Queen, Coldplay
pop: One direction, backstreet boys
indie: Arcade fire
|
|
null | false
| null |
The Thirteen Colonies, also known as the Thirteen British Colonies, the Thirteen American Colonies, were a group of British colonies on the Atlantic coast of North America. Founded in the 17th and 18th centuries, the American Enlightenment led these colonies to revolution, and declaring full independence in July 1776. With victory over Britain in the American Revolutionary War the former colonies were confirmed as the new United States of America. Just prior to declaring independence, the Thirteen Colonies in their traditional groupings were: New England (New Hampshire; Massachusetts; Rhode Island; Connecticut); Middle (New York; New Jersey; Pennsylvania; Delaware); Southern (Maryland; Virginia; North Carolina; South Carolina; and Georgia). The Thirteen Colonies came to have very similar political, constitutional, and legal systems, dominated by Protestant English-speakers. The first of these colonies was Virginia Colony in 1607, a Southern colony. While all these colonies needed to become economically viable, the founding of the New England colonies, as well as the colonies of Maryland and Pennsylvania, were substantially motivated by their founders' concerns related to the practice of religion. The other colonies were founded for business and economic expansion. The Middle Colonies were established on an earlier Dutch colony, New Netherland. All the Thirteen Colonies were part of Britain's possessions in the New World, which also included territory in Canada, Florida, and the Caribbean.
|
How many and which states were US Colonies?
|
There were 13 British Colonies that became States in the United States of America in 1776: Connecticut, Delaware, Georgia, Maryland, Massachusetts, New Hampshire, New Jersey, New York, North Carolina, Pennsylvania, Rhode Island, South Carolina, and Virginia.
|
null | false
| 183
|
Two lines of work are related to our paper: research on context-aware translation (described in Section SECREF8 ) and research on focused evaluation of pronoun translation (described in Section SECREF11 ).
Two lines of work are related to our paper: research on context-aware translation (described in Section 2.1) and research on focused evaluation of pronoun translation (described in Section 2.2).
|
What are the two lines of work that are related to this paper?
|
They are research on context-aware translation and research on focused evaluation of pronoun translation.
|
null | false
| null |
Describe the benefits of riding the bus to work instead of driving a car.
|
Riding the bus to work instead of driving a car to work reduces greenhouse gas emissions. While you are riding the bus, you can also do other tasks that you could not do while driving, such as catching up on work, messaging your friends, or taking a nap. In areas where bus lanes exist, they can make the bus as fast or faster than a car for your commute because the bus won't get stuck in traffic. Finally, if you take the bus, you don't have to worry about finding parking for your car at your destination.
|
|
null | false
| 85
|
In this paper, we have shown that using the designed RNN based network we could increase the performance of classification dramatically. We showed that keeping the sequential nature of the data can be hugely beneficial when working with textual data especially faced with the hard task of detecting more complex phenomena like emotions. We accomplished that by using a recurrent network in the process of generating our hidden representation. We have also used a max-pooling layer to capture the most relevant features and an average pooling layer to capture the text as a whole proving that we can achieve better performance by focusing on creating a more informative hidden representation. In future we can focus on improving these representations for example by using attention networks BIBREF37 , BIBREF38 to capture a more contextual representation or using language model based methods like BERT BIBREF39 that has been shown very successful in various NLP tasks.
We showed that keeping the sequential nature of the data can be hugely beneficial when working with textual data especially faced with the hard task of detecting more complex phenomena like emotions.
|
What are the results of the author's experiment?
|
Keeping the sequential nature of the data can be hugely beneficial when working with textual data especially faced with the hard task of detecting more complex phenomena like emotions.
|
null | false
| null |
What are four movies with Sandra Bullock.
|
Four movies with Sandra Bullock are The Blind Side, Bird Box, Gravity, and Miss Congeniality.
|
|
null | false
| 222
|
Negotiations, either between individuals or entities, are ubiquitous in everyday human interactions ranging from sales to legal proceedings. Being a good negotiator is a complex skill, requiring the ability to understand the partner's motives, ability to reason and to communicate effectively, making it a challenging task for an automated system. While research in building automatically negotiating agents has primarily focused on agent-agent negotiations BIBREF0, BIBREF1, there is a recent interest in agent-human negotiations BIBREF2 as well. Such agents may act as mediators or can be helpful for pedagogical purposes BIBREF3.
Efforts in agent-human negotiations involving free-form natural language as a means of communication are rather sparse. Researchers BIBREF4 recently studied natural language negotiations in buyer-seller bargaining setup, which is comparatively less restricted than previously studied game environments BIBREF5, BIBREF6. Lack of a well-defined structure in such negotiations allows humans or agents to express themselves more freely, which better emulates a realistic scenario. Interestingly, this also provides an exciting research opportunity: how can an agent leverage the behavioral cues in natural language to direct its negotiation strategies? Understanding the impact of natural language on negotiation outcomes through a data-driven neural framework is the primary objective of this work.
We focus on buyer-seller negotiations BIBREF4 where two individuals negotiate the price of a given product. Leveraging the recent advancements BIBREF7, BIBREF8 in pre-trained language encoders, we attempt to predict negotiation outcomes early on in the conversation, in a completely data-driven manner (Figure FIGREF3). Early prediction of outcomes is essential for effective planning of an automatically negotiating agent. Although there have been attempts to gain insights into negotiations BIBREF9, BIBREF10, to the best of our knowledge, we are the first to study early natural language cues through a data-driven neural system (Section SECREF3). Our evaluations show that natural language allows the models to make better predictions by looking at only a fraction of the negotiation. Rather than just realizing the strategy in natural language, our empirical results suggest that language can be crucial in the planning as well. We provide a sample negotiation from the test set BIBREF4 along with our model predictions in Table TABREF1.
We define a metric called Accuracy ± k, measuring the %age of cases where the model performs ‘well’, as controlled by k. This allows us to compare our methods in a useful manner (Section 4).
|
What is the Accuracy ± k?
|
It is metric that measures the %age of cases where the model performs ‘well’, as controlled by k.
|
null | false
| null |
Please label which of the following appear in the game Clue as suspects, weapons, or rooms: Atrium, Billiard Room, Colonel Mustard, Conservatory, General Ketchup, Lead Pipe, Mrs. Peacock, Rope
|
In the classic edition of board game Clue, Colonel Mustard and Mrs. Peacock are suspects, the Lead Pipe and Rope are weapons, and Billiard Room and Conservatory are rooms. While the Atrium and General Ketchup seem like a plausible room and suspect respectively, neither is in game.
|
|
null | false
| null |
What is the plot of Harry Potter?
|
The books concern a wizard called Harry Potter and his journey through Hogwarts School of Witchcraft and Wizardry. The stories tell of him overcoming dangerous obstacles to defeat the Dark Wizard Lord Voldemort, who killed his parents when Harry was 15 months old.
|
|
null | false
| null |
What team has won the most Stanley Cups?
|
The Montreal Canadiens have won the most Stanley Cups with 24.
|
|
null | false
| 123
|
Using GANs, we can train generative models in a two-player game setting between a discriminator and a generator, where the discriminator (a binary classifier) learns to distinguish between real and fake data samples and the generator tries to fool the discriminator by generating authentic and high quality output BIBREF16. GANs have shown to be successful in image generation tasks BIBREF17 and recently, some progress has been observed in text generation BIBREF13, BIBREF12, BIBREF15. Our generator is a language model trained using backpropagation through time BIBREF18. During the pre-training phase we optimize for MLE and during the GAN training phase, we optimize on the creativity reward from the discriminator. The discriminator's encoder has the same architecture as the generator encoder module with the addition of a pooled decoder layer. The decoder contains 3 $[Dense Batch Normalization,ReLU]$ blocks and an addtional $Sigmoid$ layer. The discriminator decoder takes the hidden state at the last time step of a sequence concatenated with both the max-pooled and mean-pooled representation of the hidden states BIBREF19 and outputs a number in the range $[0,1]$. The difficulty of using GANs in text generation comes from the discrete nature of text, making the model non-differentiable hence, we update parameters for the generator model with policy gradients as described in Yu BIBREF15.
We utilize AWD-LSTM BIBREF20 and TransformerXL BIBREF21 based language models. For model hyperparameters please to refer to Supplementary Section Table TABREF5. We use Adam optimizer BIBREF22 with $\beta 1= 0.7$ and $\beta 2= 0.8$ similar to BIBREF19 and use a batch size of 50. Other practices for LM training were the same as BIBREF21 and BIBREF20 for Transformer-XL and AWD-LSTM respectively. We refer to our proposed GAN as Creative-GAN and compare it to a baseline (a language model equivalent to our pre-trained generator) and a GumbelGAN model BIBREF14 across all proposed datasets. We use three creative English datasets with distinct linguistic characteristics: (1) A corpus of 740 classical and contemporary English poems, (2) a corpus of 14950 metaphor sentences retrieved from a metaphor database website and (3) a corpus of 1500 song lyrics ranging across genres. The mix of linguistic styles within this corpus offers the potential for interesting variation during the generation phase. We use the same pre-processing as in earlier work BIBREF19, BIBREF23. We reserve 10% of our data for test set and another 10% for our validation set.
We first pre-train our generator on the Gutenberg dataset BIBREF24 for 20 epochs and then fine-tune BIBREF19 them to our target datasets with a language modeling objective. The discriminator's encoder is initialized to the same weights as our fine-tuned language model. Once we have our fine-tuned encoders for each target dataset, we train in an adversarial manner. The discriminator objective here is to score the quality of the creative text. The discriminator is trained for 3 iterations for every iteration of the generator, a practice seen in previous work BIBREF25. Creative-GAN relies on using the reward from the discriminator BIBREF12, BIBREF15 for backpropagation. We follow a similar training procedure for GumbelGAN. Outputs are generated through sampling over a multinomial distribution for all methods, instead of $argmax$ on the log-likelihood probabilities, as sampling has shown to produce better output quality BIBREF4. Please refer to Supplementary Section Table TABREF6 for training parameters of each dataset and Table TABREF5 for hyperparameters of each encoder. We pick these values after experimentation with our validation set. Training and output generation code can be found online.
During the pre-training phase we optimize for MLE and during the GAN training phase, we optimize on the creativity reward from the discriminator. The discriminator’s encoder has the same architecture as the generator encoder module with the addition of a pooled decoder layer. The decoder contains 3 [DenseBatchNormalization, ReLU] blocks and an addtional Sigmoid layer.
|
What blocks does the pooled decoder of the discriminator’s encoder have?
|
3 [DenseBatchNormalization, ReLU] blocks.
|
null | false
| null |
What size of snowboard should I use?
|
The size of snowboard you should use is based on many different factors such as the skill of the rider, the weight and height of the rider, as well as the type of terrain you plan on riding. It is best to consult the manufacturer of the snowboard maker in order to make the best decision on the size of snowboard to use.
|
|
null | false
| null |
Should I use microservices for my project? How many microservices should I have?
|
Every microservice you add increases the testing and deployment complexity for your project. However there are a few reasons you would want to split off separate microservices:
1. the scaling requirements are different (e.g., maybe part of your code needs to be deployed to 100 servers and other parts just to 5 servers).
2. the deployment model is different (e.g., maybe part of your code needs to be sharded by customer id, and other parts not sharded)
3. they are logically different parts of the code or of the company and they can benefit from moving independently for each other.
So split off as many microservices as required by the rules above, but don't split off too many.
|
|
null | false
| null |
Name some fruits that have green skin.
|
Grapes, Green Apples and Avocados all have green skin.
|
|
null | false
| null |
A farmer is on his way to market. He's going to sell a wolf, a goat, and a cabbage. On the way, he comes to a river crossing. The river is too wide to ford, so he has to take a tiny boat across. The boat is so small that he can take at most one of the three items with him in the boat at a time. Left alone, the wolf would devour the goat, and the goat would devour the cabbage. How can the farmer safely transport all three items across the river?
|
The wolf and the cabbage can be left together with no risk to either. So the farmer first takes the goat across. He returns to the near side, retrieves the cabbage, then takes it across to the far side. Then he returns to the near side with the goat. At the near side, he leaves the goat and takes the wolf to the far side, leaving the wolf with the cabbage. Finally, he returns to the near side to take the goat across for a second time. Problem solved!
|
|
null | false
| null |
What is the difference between bourbon and whiskey?
|
While all bourbons are whiskeys, they have to be made with at least 51% corn to be classified as a bourbon.
|
|
null | false
| null |
Which of the current Girl Scout cookies is based on one of the original Girl Scout cookies?
|
Trefoils are based on the original shortbread recipe
|
|
null | false
| 11
|
People's personality and motivations are manifest in their everyday language usage. With the emergence of social media, ample examples of such usage are procurable. In this paper, we aim to analyze the vocabulary used by close to 200,000 Blogger users in the U.S. with the purpose of geographically portraying various demographic, linguistic, and psychological dimensions at the state level. We give a description of a web-based tool for viewing maps that depict various characteristics of the social media users as derived from this large blog dataset of over two billion words.
In this paper, we aim to analyze the vocabulary used by close to 200,000 Blogger users in the U.S. with the purpose of geographically portraying various demographic, linguistic, and psychological dimensions at the state level.
|
What is the purpose of this article?
|
In this paper, the author aims to analyze the vocabulary used by close to 200,000 Blogger users in the U.S. with the purpose of geographically portraying various demographic, linguistic, and psychological dimensions at the state level.
|
1709.10217
| false
| null |
In task 1, there are two top categories, namely, chit-chat and task-oriented dialogue. The task-oriented dialogue also includes 30 sub categories. In this evaluation, we only consider to classify the user intent in single utterance.
In task 1, there are two top categories, namely, chit-chat and task-oriented dialogue.
|
How many intents were classified?
|
The answers are shown as follows:
* two
|
null | false
| null |
Who are some of the best guitarists of all time?
|
Jimi Hendrix
Duane Allman
B.B. King
Eric Clapton
Robert Johnson
Chuck Berry
Stevie Ray Vaughan
Ry Cooder
Jimmy Page
Keith Richards
Kirk Hammett
Kurt Cobain
Jerry Garcia
Jeff Beck
Carlos Santana
Johnny Ramone
Jack White
John Frusciante
Richard Thompson
James Burton
George Harrison
Mike Bloomfield
Warren Haynes
The Edge
Freddie King
Tom Morello
Mark Knopfler
Stephen Stills
Ron Asheton
Buddy Guy
Dick Dale
John Cipollina
Lee Ranaldo
Thurston Moore
John Fahey
Steve Cropper
Bo Diddley
Peter Green
Brian May
John Fogerty
Clarence White
Robert Fripp
Eddie Hazel
Scotty Moore
Frank Zappa
Les Paul
T-Bone Walker
Joe Perry
John McLaughlin
Pete Townshend
Paul Kossoff
Lou Reed
Mickey Baker
Jorma Kaukonen
Ritchie Blackmore
Tom Verlaine
Roy Buchanan
Dickey Betts
Jonny Greenwood
Ed O'Brien
Ike Turner
Zoot Horn Rollo
Danny Gatton
Mick Ronson
Hubert Sumlin
Vernon Reid
Link Wray
Jerry Miller
Steve Howe
Eddie Van Halen
Lightnin' Hopkins
Joni Mitchell
Trey Anastasio
Johnny Winter
Adam Jones
Ali Farka Toure
Henry Vestine
Robbie Robertson
Cliff Gallup
Robert Quine
Derek Trucks
David Gilmour
Neil Young
Eddie Cochran
Randy Rhoads
Tony Iommi
Joan Jett
Dave Davies
D Boon
Glen Buxton
Robby Krieger
Fred "Sonic" Smith
Wayne Kramer
Bert Jansch
Kevin Shields
Angus Young
Robert Randolph
Leigh Stephens
Greg Ginn
Kim Thayil
|
|
null | false
| 332
|
As language processing tools become more prevalent concern has grown over their susceptibility to social biases and their potential to propagate bias BIBREF1, BIBREF2. Natural language training data inevitably reflects biases present in our society. For example, gender bias manifests itself in training data which features more examples of men than of women. Tools trained on such data will then exhibit or even amplify the biases BIBREF3.
Gender bias is a particularly important problem for Neural Machine Translation (NMT) into gender-inflected languages. An over-prevalence of some gendered forms in the training data leads to translations with identifiable errors BIBREF0. Translations are better for sentences involving men and for sentences containing stereotypical gender roles. For example, mentions of male doctors are more reliably translated than those of male nurses BIBREF2, BIBREF4.
Recent approaches to the bias problem in NLP have involved training from scratch on artificially gender-balanced versions of the original dataset BIBREF5, BIBREF6 or with de-biased embeddings BIBREF7, BIBREF8. While these approaches may be effective, training from scratch is inefficient and gender-balancing embeddings or large parallel datasets are challenging problems BIBREF9.
Instead we propose treating gender debiasing as a domain adaptation problem, since NMT models can very quickly adapt to a new domain BIBREF10. To the best of our knowledge this work is the first to attempt NMT bias reduction by fine-tuning, rather than retraining. We consider three aspects of this adaptation problem: creating less biased adaptation data, parameter adaptation using this data, and inference with the debiased models produced by adaptation.
Regarding data, we suggest that a small, trusted gender-balanced set could allow more efficient and effective gender debiasing than a larger, noisier set. To explore this we create a tiny, handcrafted profession-based dataset for transfer learning. For contrast, we also consider fine-tuning on a counterfactual subset of the full dataset and propose a straightforward scheme for artificially gender-balancing parallel text for NMT.
We find that during domain adaptation improvement on the gender-debiased domain comes at the expense of translation quality due to catastrophic forgetting BIBREF11. We can balance improvement and forgetting with a regularised training procedure, Elastic Weight Consolidation (EWC), or in inference by a two-step lattice rescoring procedure.
We experiment with three language pairs, assessing the impact of debiasing on general domain BLEU and on the WinoMT challenge set BIBREF0. We find that continued training on the handcrafted set gives far stronger and more consistent improvements in gender-debiasing with orders of magnitude less training time, although as expected general translation performance as measured by BLEU decreases.
We further show that regularised adaptation with EWC can reduce bias while limiting degradation in general translation quality. We also present a lattice rescoring procedure in which initial hypotheses produced by the biased baseline system are transduced to create gender-inflected search spaces which can be rescored by the adapted model. We believe this approach, rescoring with models targeted to remove bias, is novel in NMT. The rescoring procedure improves WinoMT accuracy by up to 30% with no decrease in BLEU on the general test set.
Recent recommendations for ethics in Artificial Intelligence have suggested that social biases or imbalances in a dataset be addressed prior to model training BIBREF12. This recommendation presupposes that the source of bias in a dataset is both obvious and easily adjusted. We show that debiasing a full NMT dataset is difficult, and suggest alternative efficient and effective approaches for debiasing a model after it is trained. This avoids the need to identify and remove all possible biases prior to training, and has the added benefit of preserving privacy, since no access to the original data or knowledge of its contents is required. As evidence, in section SECREF43, we show this scheme can be applied to remove gender bias in the output of ‘black box‘ online commercial MT systems.
Gender bias is a particularly important problem for Neural Machine Translation (NMT) into gender-inflected languages.
|
Is the Gender bias a particularly important problem for Neural Machine Translation (NMT) into gender-inflected languages?
|
Yes, it is.
|
null | false
| null |
Write a haiku about cats
|
Whiskers fluttering
A silent leap, and then: crash
My coffee mug breaks.
|
|
null | false
| null |
A druid was a member of the high-ranking class in ancient Celtic cultures. Druids were religious leaders as well as legal authorities, adjudicators, lorekeepers, medical professionals and political advisors. Druids left no written accounts. While they were reported to have been literate, they are believed to have been prevented by doctrine from recording their knowledge in written form. Their beliefs and practices are attested in some detail by their contemporaries from other cultures, such as the Romans and the Greeks.
|
What is a Druid?
|
A druid was a member of the high-ranking class in ancient Celtic cultures.
|
null | false
| null |
When was the Victorian Football League renamed to the Australian Football League?
|
1990
|
|
null | false
| 160
|
The purpose of text classification is that, given a sentence $x$ , the model should predict labels $\hat{y}$ from a pre-defined label set $\mathcal {Y}$ . From the description in the previous section, we can compute the distributed representation $\mathbf {h}_j$ of the phrase at node $j$ of a tree:
$$\mathbf {h}_{j} &= \operatornamewithlimits{DC-TreeNN}(\mathbf {x}_j, \mathbf {h}_{j}^l, \mathbf {h}_{j}^r,\theta )$$ (Eq. 18)
After this recursive process, the hidden state $\mathbf {h}_R$ at the root node is used as the sentential representation, which then followed by a softmax classifier to predict the probability distribution over classes.
$${\hat{\mathbf {y}}} = \operatornamewithlimits{softmax}(\mathbf {W}_t \mathbf {h}_R + \mathbf {b}_t)$$ (Eq. 19)
where ${\hat{\mathbf {y}}}$ is prediction probabilities, $\mathbf {W}_t$ and $\mathbf {b}_t$ are the parameters of the classifier.
We evaluate our models on five different datasets. The detailed statistics about the five datasets are listed in Table 2 . Each dataset is briefly described as follows.
SST The movie reviews with two classes (negative, positive) in the Stanford Sentiment Treebank BIBREF23 .
MR The movie reviews with two classes BIBREF24 .
QC The TREC questions dataset involves six different question types. BIBREF25 .
SUBJ Subjectivity dataset where the goal is to classify each instance (snippet) as being subjective or objective. BIBREF26
IE Idiom enhanced sentiment classification. BIBREF27 . Each sentence contains at least one idiom.
As shown in Table 3 , DC-TreeLSTM consistently outperforms RecNN, MV-RecNN, RNTN, and TreeLSTM by a large margin while achieving comparable results to the CNN and using much fewer parameters.(The number of parameters in our models is approximately 10K while in CNN the number of parameters is about 400K). Compared with RecNN, DC-RecNN performs better, indicating the effectiveness of the dynamic compositional function. Additionally, both DC-RecNN and DC-TreeLSTM achieve substantial improvement on IE dataset, which covers the richness of compositionality (idiomaticity). We attribute the success on IE to its power in modeling more complicated compositionality.
• SST The movie reviews with two classes (negative, positive) in the Stanford Sentiment Treebank [Socher et al., 2013b]. • MR The movie reviews with two classes [Pang and Lee, 2005]. • QC The TREC questions dataset involves six different question types. [Li and Roth, 2002]. • SUBJ Subjectivity dataset where the goal is to classify each instance (snippet) as being subjective or objective. [Pang and Lee, 2004] • IE Idiom enhanced sentiment classification. [Williams et al., 2015]. Each sentence contains at least one idiom.
|
What data sets did the authors use to evaluate their model?
|
• SST The movie reviews with two classes (negative, positive) in the Stanford Sentiment Treebank. • MR The movie reviews with two classes. • QC The TREC questions dataset involves six different question types. • SUBJ Subjectivity dataset where the goal is to classify each instance (snippet) as being subjective or objective. • IE Idiom enhanced sentiment classification.
|
null | false
| null |
Who was favored to win the 2023 NCAA Men's Collegiate Basketball Championship game?
|
The final odds from the primary online gambling sites in the 2023 NCAA Men's Collegiate Basketball Championship game gave the University of Connecticut (UCONN) a 7.5 point advantage over San Diego State University (SDSU).
|
|
null | false
| null |
What are the four great inventions of Ancient China?
|
The four great inventions of ancient China are paper-making, printing, gunpowder and compass. Paper that was most similar to modern materials, was first invented by a court official named Cai Lun in the early 2nd century China. It was relatively cheap, light and suitable for brush writing, and was produced using bark, rags, wheat stalks and other materials. As for printing, the first book with a verifiable date of printing was a Buddhist sutra that was printed in the Tang Dynasty, around the year of 868. The use of gunpowder technique was first discovered by Chinese monks in the 9th century CE, during their quest for a life-extending elixir. Last but not least, one of the first books to describe the magnetic compass was written by Shen Kuo in the Song Dynasty, 11th Century CE.
|
|
null | false
| null |
He enrolled in University of California, Davis (UC Davis) initially studying biology, followed by anthropology, and ending with a focus on fine art. While attending UC Davis, Gilhooly served as artist Robert Arneson's assistant starting in 1963. He graduated from UC Davis with a BA degree in 1965, and an MA degree in 1967.
|
What college did David Gilhooly go to?
|
David Gilhooly went to the University of California, Davis (UC Davis).
|
null | false
| null |
What is a vegetarian diet?
|
A vegetarian diet is one in which someone eats no meat, but will consume other animal products such as eggs or dairy
|
|
null | false
| 50
|
Word embeddings were first introduced in BIBREF2 , BIBREF3 and there have been many variants BIBREF4 , BIBREF5 , BIBREF6 , BIBREF7 . Common for all of them is the idea that words can be represented by latent feature vectors. These feature vectors are optimized to maximize the conditional probability of the dataset. Recently BIBREF1 extended the idea of word embeddings to other types of data. EqEmb expand the idea of word embeddings to a new type of data points – equations.
There have been different proposed approaches for representing mathematical equations. BIBREF8 introduced the symbol layout tree, a representation that encodes the spatial relationship of variables and operators for the purpose of indexing and retrieving mathematical equations. Our work also falls into the framework of mathematical language processing (MLP) BIBREF9 whose first step is converting mathematical solutions into a series of numerical features.
There have been different proposed approaches for representing mathematical equations. Zanibbi et al. (2016) introduced the symbol layout tree, a representation that encodes the spatial relationship of variables and operators for the purpose of indexing and retrieving mathematical equations. Our work also falls into the framework of mathematical language processing (MLP) (Lan et al., 2015) whose first step is converting mathematical solutions into a series of numerical features.
|
What is the symbol layout tree introduced by Zanibbi et al.?
|
A representation that encodes the spatial relationship of variables and operators for the purpose of indexing and retrieving mathematical equations.
|
null | false
| 446
|
Table 9: Comparison of FedBYOL without exponential moving average (EMA) and stop-gradient (sg) on the CIFAR datasets. FedBYOL w/o EMA and sg can hardly learn, but updating both Wk and Wtk with Wg (update-both) enables it to achieve comparable results.
|
FL clients should not be stateful, do you agree?
|
Thank you for raising the concern. Actually, this paper focuses on cross-silo FL rather than cross-device FL and we have revised the manuscript to clarify this point. We have also refined Algorithm 1 to handle client sampling on every training round. Specifically, when clients are newly sampled, we update their online and target encoders with the global model; when clients are sampled in the previous round, we update only the online encoders with EMA. In definition, cross-device FL assumes there are millions of stateless clients that might participate in training just once. While due to the constraints of experimental settings, the majority of studies conduct experiments with at most hundreds of clients. FedEMA can work under such experimental settings by caching states of clients in the server. When the number of clients scales to millions, FedEMA degrades to FedBYOL that updates both encoders --- without keeping any local states. We have included the discussion on cross-silo FL and cross-device FL in Appendix B.3.
|
1908.07822
| false
| null |
The first five methods are the most common class (MCC), $KLD$, $LS \cup KLD$, $LS \cup KLD \cup CONN$, and $KLD \cup LS \cup LS_{inter}$. $KLD$, $LS$, and $CONN$ represent KL-divergence score, lexical semantic feature, and categorical feature respectively. These methods are used as baselines in Hidey et al.'s work. $KLD$ and $LS \cup KLD$ acquire the best accuracy and precision on the Training set. $LS \cup KLD \cup CONN$ and $KLD \cup LS \cup LS_{inter}$ are the best systems with the highest recall and F1-score respectively. The next five are the most commonly used methods in text classification. They are TextCNN, TextRNN, SASE, DPCNN, and BERT. In our experiment, we reproduced all of them except BERT. For BERT, we use the public released pre-trained language model (base). and fine-tuned it on each dataset. The detailed information about these baselines is listed as follows:
$LS \cup KLD \cup CONN$ and $KLD \cup LS \cup LS_{inter}$ are the best systems with the highest recall and F1-score respectively.
|
What was previous state-of-the-art approach?
|
The answers are shown as follows:
* TextCNN, TextRNN, SASE, DPCNN, and BERT
* $LS \cup KLD \cup CONN$ and $KLD \cup LS \cup LS_{inter}$ are the best systems with the highest recall and F1-score respectively
|
null | false
| null |
Artificial intelligence (AI) is intelligence demonstrated by machines, as opposed to intelligence of humans and other animals. Example tasks in which this is done include speech recognition, computer vision, translation between (natural) languages, as well as other mappings of inputs.
|
What are some examples of tasks performed by AI?
|
Examples of tasks performed by AI include speech recognition, computer vision, translation between (natural) languages, and other mappings of inputs.
|
null | false
| 410
|
4 methods, i.e., GCQL (Ours), CQL, REDQ and AWAC, are tested on 3 tasks, i.e., Walker, Hopper and HalfCheetah, and each task has 5 different kinds of offline dataset, which are random-v0, medium-v0, expert-v0, medium-expert-v0, and medium-replay-v0. As online interactions are available, we take the maximization over testing scores during the whole training process, and report the average of these max scores across five different seeds in Table. Please note that the results of AWAC are directly taken from their paper. As shown in Table, our method gains a better performance than the baselines in all tasks. Particularly, for the medium-replay and random dataset, our GCQL outperforms the baselines by a large margin. By contrast, CQL also achieves an expert performance on the high-quality dataset, e.g., expert and medium-expert tasks. For these high-quality datasets, our method achieves a comparable or slightly better performance over CQL. In terms of the online RL method REDQ, it fails on almost all tasks. Even though we employ an online limited exploration process, conventional online RL cannot benefit from such limited experiences. On the contrary, such limited online experiences could bring catastrophic consequences, e.g., on the halfcheetah-medium-replay task, REDQ suffers a significant instability issue. On the other hand, when dataset quality is high, e.g., expert or medium-expert dataset, our method can leverage the benefit of the offline RL that gains an expert performance with very limited interaction steps, such as around 20K steps. By contrast, when dataset quality is poor, e.g., the random or medium-replay dataset, only our method can learn effectively with less than 100K online steps, which demonstrate that our method can take the advantage of the limited online experiences as much as possible. In sum, our method can benefit the most from the limited online experience while still maintain the offline learning's ability. Besides, We also include the corresponding learning curves in Appendix A for a comprehensive understanding. These learning curves demonstrate that our method also achieves a better final average score at the last iteration in most cases.
As online interactions are available, we take the maximization over testing scores during the whole training process, and report the average of these max scores across five different seeds in Table 1.
|
"we report the max average score as performance", this is ambiguous. As I see it, we have different points along the x-axis and different seeds. What are taking the maximum over? What are we averaging over?"
|
The maximum is the max score for each seed. and we average across different seeds. We have updated it accordingly in section 5.2.
|
null | false
| 139
|
It is well known that language has certain structural properties which allows natural language speakers to make “infinite use of finite means" BIBREF3 . This structure allows us to generalize beyond the typical machine learning definition of generalization BIBREF4 (which considers performance on the distribution that generated the training set), permitting the understanding of any utterance sharing the same structure, regardless of probability. For example, sentences of length 100 typically do not appear in natural text or speech (our personal 'training set'), but can be understood regardless due to their structure. We refer to this notion as linguistic generalization .
Many problems in NLP are treated as sequence to sequence tasks with solutions built on seq2seq-attention based models. While these models perform very well on standard datasets and also appear to capture some linguistic structure BIBREF5 , BIBREF6 , BIBREF7 , they also can be quite brittle, typically breaking on uncharacteristic inputs BIBREF8 , BIBREF1 , indicating that the extent of linguistic generalization these models achieve is still somewhat lacking.
Due to the high capacity of these models, it is not unreasonable to expect them to learn some structure from the data. However, learning structure is not a sufficient condition to achieving linguistic generalization. If this structure is to be usable on data outside the training distribution, the model must learn the structure without additionally learning (overfitting on) patterns specific to the training data. One may hope, given the right hyperparameter configuration and regularization, that a model converges to a solution that captures the reusable structure without overfitting too much on the training set. While this solution exists in theory, in practice, it may be difficult to find.
In this work, we look at the feasibility of training and tuning seq2seq-attention models towards a solution that generalizes in this linguistic sense. In particular, we train models on a symbol replacement task with a well defined generalizable structure. The task is simple enough that all models achieve near perfect accuracy on the standard test set, i.e., where the inputs are drawn from the same distribution as that of the training set. We then test these models for linguistic generalization by creating test sets of uncharacteristic inputs, i.e., inputs that are not typical in the training distribution but still solvable given that the generalizable structure was learned. Our results indicate that generalization is highly sensitive; such that even changes in the random seed can drastically affect the ability to generalize. This dependence on an element that is not (or ideally should not be) a hyperparameter suggests that the line between generalization and failure is quite fine, and may not be feasible to reach simply by hyperparameter tuning alone.
In particular, we train models on a symbol replacement task with a well defined generalizable structure.
|
How do the authors train the models in this paper?
|
The authors train models on a symbol replacement task with a well defined generalizable structure.
|
null | false
| null |
Identify which instrument is string or percussion: Rototom, Mugni
|
Mugni is string, Rototom is percussion.
|
|
1907.00854
| false
| null |
The current release of Katecheo uses a Bi-Directional Attention Flow, or BiDAF, model for reading comprehension BIBREF6 . This BiDAF model includes a Convolutional Neural Network (CNN) based character level embedding layer, a word embedding layer that uses pre-trained GloVE embeddings, a Long Short-Term Memory Network (LSTM) based contextual embedding layer, an “attention flow layer", and a modeling layer include bi-directional LSTMs. We are using a pre-trained version of BiDAF available in the AllenNLP BIBREF7 library.
Future releases of Katecheo will include the ability to swap out the reading comprehension model for newer architectures based on, e.g., BERT BIBREF8 or XLNet BIBREF9 or custom trained models.
Architecture and Configuration
The current release of Katecheo uses a Bi-Directional Attention Flow, or BiDAF, model for reading comprehension BIBREF6 .
Future releases of Katecheo will include the ability to swap out the reading comprehension model for newer architectures based on, e.g., BERT BIBREF8 or XLNet BIBREF9 or custom trained models.
Architecture and Configuration
|
what pretrained models were used?
|
The answers are shown as follows:
* BiDAF
* BERT
|
null | false
| null |
Most water vapors found mostly in the ocean returns to it, but winds carry water vapor over land at the same rate as runoff into the sea, about 47 Tt per year whilst evaporation and transpiration happening in land masses also contribute another 72 Tt per year. Precipitation, at a rate of 119 Tt per year over land, has several forms: most commonly rain, snow, and hail, with some contribution from fog and dew. Dew is small drops of water that are condensed when a high density of water vapor meets a cool surface. Dew usually forms in the morning when the temperature is the lowest, just before sunrise and when the temperature of the earth's surface starts to increase. Condensed water in the air may also refract sunlight to produce rainbows.
|
Given the following paragraph on the water cycle, what forms of precipitation do exist?
|
Water can precipitate in the form of rain, snow, and hail as well as in the form of fog and dew. Rain is the most common form of precipitation.
|
1908.10422
| false
| null |
Ensemble, which selects a sentence using 100 agents trained on clustered dialogues as described in section SECREF4 – the agent in focus is chosen using a regressor as predictor of dialogue reward INLINEFORM0 using a similar neural net as the ChatDQN agents except for the final layer having one node and using Batch Normalisation BIBREF44 between hidden layers as in BIBREF36 ;
Ensemble, which selects a sentence using 100 agents trained on clustered dialogues as described in section SECREF4 – the agent in focus is chosen using a regressor as predictor of dialogue reward INLINEFORM0 using a similar neural net as the ChatDQN agents except for the final layer having one node and using Batch Normalisation BIBREF44 between hidden layers as in BIBREF36 ;
|
How many agents do they ensemble over?
|
The answers are shown as follows:
* 100
|
null | false
| null |
What do you think a conversation between two AI bots would look like?
|
I do not know but I would like to imagine the two bots are both trained on enormous corpse of data available online. If the datasets are different, the conversation could go very interesting and unexpected for humans ways. The common ML algorithms might be similar or shared so I wonder if the reasoning behind the bots will be similar and a pattern could be drawn on them.
|
|
null | false
| null |
What is the difference between American Football and Futbol?
|
In America, football refers to the NFL style of football. This involves heavily padded players, scripted plays and a game where the goal is to get the ball into an endzone with a mix of pass and run plays. International Futbol refers to the game that is played primarily with a players feet, passing the ball and shooting with only their feet of head. Players in Futbol are not allowed to use their hands unless they are the goalie defending the goal. The goal of Futbol is to get the ball into the opposing team's net.
|
|
null | false
| null |
What is the Nehemiah Hubbard House
|
The Nehemiah Hubbard House is a historic house at the corner of Laurel Grove and Wadsworth Street, Middletown, Connecticut. Built in 1745, it is a center-chimney colonial style house built of clapboard siding and brownstone foundation with wood shingle roof; using a structural system of wood frame, post and beam with gable roof. It was built as a residence which is its current use.
|
|
null | false
| null |
Imagine you are the CEO of Tiktok and have to give an argument in front of the Senate on why the US should not ban Tiktok.
|
Hello members of the Senate. My name is Shou Zi Chew, and I am the CEO of Tiktok. Tiktok is a platform that surfaces short videos based on an algorithm that predicts your interests and likelihood of engaging with a video. We have a very large user base who, through engaging with our platform, have found communities that support and provide vast amounts of new knowledge. Not only is Tiktok a source of information and news, we have invested in making sure only appropriate content is surfaced, especially for children. We take privacy and security very seriously. Some members of the Senate are concerned that because Tiktok is held by a Chinese company, that we have secret ties to China. I can assure you that that is not the case. I believe Tiktok has brought users entertainment and education that warrants it to continue operating as a company. Please consider supporting Tiktok. Thank you member of the Senate for taking the time today to learn about Tiktok.
|
|
null | false
| 5
|
We train the LangID and NoLangID versions of our model each on three subsets of the Wiktionary data:
LangID-High and NoLangID-High: Trained on data from the 85 languages for which BIBREF13 used non-adapted wFST models.
LangID-Adapted and NoLangID-Adapted: Trained on data from any of the 229 languages for which they built adapted models. Because many of these languages had no training data at all, the model is actually only trained on data in 157 languages. As is noted above, the Adapted set omits 23 languages which are in the High test set.
LangID-All and NoLangID-All: Trained on data in all 311 languages in the Wiktionary training corpus.
In order to ease comparison to Deri and Knight's system, we limited our use of the training corpus to 10,000 words per language. We set aside 10 percent of the data in each language for validation, so the maximum number of training words for any language is 9000 for our systems.
We set aside 10 percent of the data in each language for validation, so the maximum number of training words for any language is 9000 for our systems.
|
What is the training-validation ratio？
|
9:1.
|
null | false
| null |
Come up with the best way to build a sales pipeline
|
There are a number of ways that a sales representative can build up their sales pipeline, or funnel. What it comes down to is that you need to have the activity levels that will support having a minimum of a 3:1 pipeline to goal ratio. This can be done through cold calling, LinkedIN messaging, referrals, or email, etc
|
|
null | false
| null |
When is the best time of day to exercise?
|
The best time of day to exercise is the time you feel is best for you! Many people find working out in the morning energizes them for the rest of the day, and they feel the most willing to start their day off with rigorous activity. Others find a midday workout to revitalize them when hitting a slump. Some folks are more inclined to workout in the evenings to release the stress of the day. The time of day is less important than the type of activity, ensuring the activity is safe for your health, and that you are properly fueled with healthy nutrition and well rested. Daily activity is important to overall physical and mental health, so even light to moderate exercise has tremendous benefits. In short, worry less about the time of day and more on the quality of the activity and benefits you get.
|
|
null | false
| null |
What is ABAP?
|
A programming language for developing applications for the SAP
|
|
1810.01570
| false
| null |
As a benchmark, we will use the results of the systems by Burckhardt et al. BIBREF22 , Liu et al. BIBREF18 , Dernoncourt et al. BIBREF9 and Yang et al. BIBREF10 on the i2b2 dataset and the performance of Burckhardt et al. on the nursing corpus. Note that Burckhardt et al. used the entire data set for their results as it is an unsupervised learning system while we had to split our data set into 60% training data and 40% testing data.
As a benchmark, we will use the results of the systems by Burckhardt et al. BIBREF22 , Liu et al. BIBREF18 , Dernoncourt et al. BIBREF9 and Yang et al. BIBREF10 on the i2b2 dataset and the performance of Burckhardt et al. on the nursing corpus.
|
What is their baseline?
|
The answers are shown as follows:
* Burckhardt et al. BIBREF22
* Liu et al. BIBREF18
* Dernoncourt et al. BIBREF9
* Yang et al. BIBREF10
|
null | false
| null |
The Treaty of Dunkirk was signed by France and the United Kingdom on 4 March 1947, during the aftermath of World War II and the start of the Cold War, as a Treaty of Alliance and Mutual Assistance in the event of possible attacks by Germany or the Soviet Union. In March 1948, this alliance was expanded in the Treaty of Brussels to include the Benelux countries, forming the Brussels Treaty Organization, commonly known as the Western Union. Talks for a wider military alliance, which could include North America, also began that month in the United States, where their foreign policy under the Truman Doctrine promoted international solidarity against actions they saw as communist aggression, such as the February 1948 coup d'état in Czechoslovakia. These talks resulted in the signature of the North Atlantic Treaty on 4 April 1949 by the member states of the Western Union plus the United States, Canada, Portugal, Italy, Norway, Denmark, and Iceland. Canadian diplomat Lester B. Pearson was a key author and drafter of the treaty.
A long rectangular room with multiple rows of seated individuals on each side, and flags hanging at the far end.
West Germany joined NATO in 1955, which led to the formation of the rival Warsaw Pact during the Cold War.
The North Atlantic Treaty was largely dormant until the Korean War initiated the establishment of NATO to implement it with an integrated military structure. This included the formation of Supreme Headquarters Allied Powers Europe (SHAPE) in 1951, which adopted many of the Western Union's military structures and plans, including their agreements on standardizing equipment and agreements on stationing foreign military forces in European countries. In 1952, the post of Secretary General of NATO was established as the organization's chief civilian. That year also saw the first major NATO maritime exercises, Exercise Mainbrace and the accession of Greece and Turkey to the organization. Following the London and Paris Conferences, West Germany was permitted to rearm militarily, as they joined NATO in May 1955, which was, in turn, a major factor in the creation of the Soviet-dominated Warsaw Pact, delineating the two opposing sides of the Cold War.
The building of the Berlin Wall in 1961 marked a height in Cold War tensions, when 400,000 US troops were stationed in Europe. Doubts over the strength of the relationship between the European states and the United States ebbed and flowed, along with doubts over the credibility of the NATO defence against a prospective Soviet invasion – doubts that led to the development of the independent French nuclear deterrent and the withdrawal of France from NATO's military structure in 1966. In 1982, the newly democratic Spain joined the alliance.
The Revolutions of 1989 in Europe led to a strategic re-evaluation of NATO's purpose, nature, tasks, and focus on the continent. In October 1990, East Germany became part of the Federal Republic of Germany and the alliance, and in November 1990, the alliance signed the Treaty on Conventional Armed Forces in Europe (CFE) in Paris with the Soviet Union. It mandated specific military reductions across the continent, which continued after the collapse of the Warsaw Pact in February 1991 and the dissolution of the Soviet Union that December, which removed the de facto main adversaries of NATO. This began a draw-down of military spending and equipment in Europe. The CFE treaty allowed signatories to remove 52,000 pieces of conventional armaments in the following sixteen years, and allowed military spending by NATO's European members to decline by 28 percent from 1990 to 2015. In 1990 assurances were given by several Western leaders to Mikhail Gorbachev that NATO would not expand further east, as revealed by memoranda of private conversations. However, the final text of the Treaty on the Final Settlement with Respect to Germany, signed later that year, contained no mention of the issue of eastward expansion.
Two tall concrete sections of a wall in a grass lawn in front of a curved building with blue-glass windows.
The Fall of the Berlin Wall in 1989 marked a turning point in NATO's role in Europe, and this section of the wall is now displayed outside NATO headquarters.
In the 1990s, the organization extended its activities into political and humanitarian situations that had not formerly been NATO concerns. During the Breakup of Yugoslavia, the organization conducted its first military interventions in Bosnia from 1992 to 1995 and later Yugoslavia in 1999. These conflicts motivated a major post-Cold War military restructuring. NATO's military structure was cut back and reorganized, with new forces such as the Headquarters Allied Command Europe Rapid Reaction Corps established.
Politically, the organization sought better relations with the newly autonomous Central and Eastern European states, and diplomatic forums for regional cooperation between NATO and its neighbours were set up during this post-Cold War period, including the Partnership for Peace and the Mediterranean Dialogue initiative in 1994, the Euro-Atlantic Partnership Council in 1997, and the NATO–Russia Permanent Joint Council in 1998. At the 1999 Washington summit, Hungary, Poland, and the Czech Republic officially joined NATO, and the organization also issued new guidelines for membership with individualized "Membership Action Plans". These plans governed the subsequent addition of new alliance members. The election of French President Nicolas Sarkozy in 2007 led to a major reform of France's military position, culminating with the return to full membership on 4 April 2009, which also included France rejoining the NATO Military Command Structure, while maintaining an independent nuclear deterrent.
Article 5 of the North Atlantic treaty, requiring member states to come to the aid of any member state subject to an armed attack, was invoked for the first and only time after the September 11 attacks, after which troops were deployed to Afghanistan under the NATO-led ISAF. The organization has operated a range of additional roles since then, including sending trainers to Iraq, assisting in counter-piracy operations, and in 2011 enforcing a no-fly zone over Libya in accordance with UN Security Council Resolution 1973.
Russia's annexation of Crimea led to strong condemnation by all NATO members, and was one of the seven times that Article 4, which calls for consultation among NATO members, has been invoked. Prior times included during the Iraq War and Syrian Civil War. At the 2014 Wales summit, the leaders of NATO's member states formally committed for the first time to spend the equivalent of at least two percent of their gross domestic products on defence by 2024, which had previously been only an informal guideline. At the 2016 Warsaw summit, NATO countries agreed on the creation of NATO Enhanced Forward Presence, which deployed four multinational battalion-sized battlegroups in Estonia, Latvia, Lithuania, and Poland. Before and during the 2022 Russian invasion of Ukraine, several NATO countries sent ground troops, warships and fighter aircraft to reinforce the alliance's eastern flank, and multiple countries again invoked Article 4. In March 2022, NATO leaders met at Brussels for an extraordinary summit which also involved Group of Seven and European Union leaders. NATO member states agreed to establish four additional battlegroups in Bulgaria, Hungary, Romania, and Slovakia, and elements of the NATO Response Force were activated for the first time in NATO's history.
As of June 2022, NATO had deployed 40,000 troops along its 2,500-kilometre-long (1,550 mi) Eastern flank to deter Russian aggression. More than half of this number have been deployed in Bulgaria, Romania, Hungary, Slovakia, and Poland, which five countries muster a considerable combined ex-NATO force of 259,000 troops. To supplement Bulgaria's Air Force, Spain sent Eurofighter Typhoons, the Netherlands sent eight F-35 attack aircraft, and additional French and US attack aircraft would arrive soon as well.
NATO enjoys public support across its member states.
|
Summarize the history of NATO
|
The North Atlantic Treaty Organization (NATO, French: Organisation du traité de l'Atlantique nord), or North Atlantic Alliance, the Atlantic Alliance, the Western Alliance, is a military alliance. It was established by the North Atlantic Treaty in 1949 and it was signed in Washington, DC, United States, on April 4, 1949. Its members in 1949 were the United States, Belgium, the Netherlands, Luxembourg, France, the United Kingdom, Canada, Portugal, Italy, Norway, Denmark and Iceland. On 18 February 1952, Greece and Turkey also joined.
When West Germany joined the organization on 9 May 1955 it was described as "a decisive turning point in the history of our continent" by Halvard Lange, Foreign Minister of Norway at the time., the result was the Warsaw Pact, signed on 14 May 1955 by the Soviet Union and its satellite states as response to NATO.
After the Cold War in 1999 three former communist countries, Hungary, the Czech Republic, and Poland joined NATO. On 29 March 2004 seven more Northern European and Eastern European countries joined NATO: Estonia, Latvia and Lithuania and also Slovenia, Slovakia, Bulgaria, and Romania.
Croatia and Albania received NATO membership invitation on 3 April 2008. The Republic of Macedonia received the only conditional invitation because it was vetoed by Greece because of its name dispute with Greece.
Montenegro joined on 5 June 2017. Having changed its name to end the dispute, North Macedonia joined NATO on 27 March 2020 and became its 30th member. Ireland officially joined NATO on 8 September 2020 as an observer.
In April 2023, Finland joined NATO, following efforts to join since the 2022 Russian invasion of Ukraine.
|
null | false
| null |
How do solar panels create energy?
|
Solar panels convert the radiant energy from sunlight into electric energy in the form of direct current electricity.
|
|
null | false
| null |
What is Rust?
|
Rust is a multi-paradigm, high-level, general-purpose programming language. Rust emphasizes performance, type safety, and concurrency. Rust enforces memory safety—that is, that all references point to valid memory—without requiring the use of a garbage collector or reference counting present in other memory-safe languages. To simultaneously enforce memory safety and prevent concurrent data races, Rust's "borrow checker" tracks the object lifetime of all references in a program during compilation. Rust is popular for systems programming but also offers high-level features including some functional programming constructs.
|
|
null | false
| null |
Which of the following does not fit with the others on the list? Please explain your reasoning.
Galaxy M81
Black Eye Galaxy
Pinwheel Galaxy
Andromeda Galaxy
Sombrero Galaxy
|
Of the 5 galaxies on the list, the Sombrero Galaxy is the only one that is not clearly a spiral galaxy. While some sources do describe the Sombrero Galaxy as a spiral galaxy, most acknowledge that it has an usual shape that differs from most spiral galaxies.
|
|
null | false
| null |
Classify each of these as ground transportation or aviation transportation: airplane, truck, bicycle, unicycle, jet, wagon, helicopter.
|
Ground transportation would include all vehicles that travel along the ground. From the list you provided these include: truck, bicycle, unicycle, and wagon.
Aviation transportation includes any vehicle that flies. From the list you provided these include: airplane, jet, and helicopter.
|
|
null | false
| 154
|
All corpora provide datasets/splits for answer selection, whereas only (WikiQA, SQuAD) and (WikiQA, SelQA) provide datasets for answer extraction and answer triggering, respectively. SQuAD is much larger in size although questions in this corpus are often paraphrased multiple times. On the contrary, SQuAD's average candidates per question ( INLINEFORM0 ) is the smallest because SQuAD extracts answer candidates from paragraphs whereas the others extract them from sections or infoboxes that consist of bigger contexts. Although InfoboxQA is larger than WikiQA or SelQA, the number of token types ( INLINEFORM1 ) in InfoboxQA is smaller than those two, due to the repetitive nature of infoboxes.
All corpora show similar average answer candidate lengths ( INLINEFORM0 ), except for InfoboxQA where each line in the infobox is considered a candidate. SelQA and SQuAD show similar average question lengths ( INLINEFORM1 ) because of the similarity between their annotation schemes. It is not surprising that WikiQA's average question length is the smallest, considering their questions are taken from search queries. InfoboxQA's average question length is relatively small, due to the restricted information that can be asked from the infoboxes. InfoboxQA and WikiQA show the least question-answer word overlaps over questions and answers ( INLINEFORM2 and INLINEFORM3 in Table TABREF2 ), respectively. In terms of the F1-score for overlapping words ( INLINEFORM4 ), SQuAD gives the least portion of overlaps between question-answer pairs although WikiQA comes very close.
Fig. FIGREF4 shows the distributions of seven question types grouped deterministically from the lexicons. Although these corpora have been independently developed, a general trend is found, where the what question type dominates, followed by how and who, followed by when and where, and so on.
Fig. FIGREF6 shows the distributions of answer categories automatically classified by our Convolutional Neural Network model trained on the data distributed by li:02a. Interestingly, each corpus focuses on different categories, Numeric for WikiQA and SelQA, Entity for SQuAD, and Person for InfoboxQA, which gives enough diversities for statistical learning to build robust models.
All corpora show similar average answer candidate lengths (µc), except for INFOBOXQA where each line in the infobox is considered a candidate.
|
Do all corpora show similar average answer candidate lengths (µc)?
|
No, they don't.
|
null | false
| null |
Life arose from the Earth's first ocean, which formed some 3.8 billion years ago. Since then, water continues to be the most abundant molecule in every organism. Water is important to life because it is an effective solvent, capable of dissolving solutes such as sodium and chloride ions or other small molecules to form an aqueous solution. Once dissolved in water, these solutes are more likely to come in contact with one another and therefore take part in chemical reactions that sustain life. In terms of its molecular structure, water is a small polar molecule with a bent shape formed by the polar covalent bonds of two hydrogen (H) atoms to one oxygen (O) atom (H2O). Because the O–H bonds are polar, the oxygen atom has a slight negative charge and the two hydrogen atoms have a slight positive charge. This polar property of water allows it to attract other water molecules via hydrogen bonds, which makes water cohesive. Surface tension results from the cohesive force due to the attraction between molecules at the surface of the liquid. Water is also adhesive as it is able to adhere to the surface of any polar or charged non-water molecules. Water is denser as a liquid than it is as a solid (or ice). This unique property of water allows ice to float above liquid water such as ponds, lakes, and oceans, thereby insulating the liquid below from the cold air above. Water has the capacity to absorb energy, giving it a higher specific heat capacity than other solvents such as ethanol. Thus, a large amount of energy is needed to break the hydrogen bonds between water molecules to convert liquid water into water vapor. As a molecule, water is not completely stable as each water molecule continuously dissociates into hydrogen and hydroxyl ions before reforming into a water molecule again. In pure water, the number of hydrogen ions balances (or equals) the number of hydroxyl ions, resulting in a pH that is neutral.
|
Based on the following paragraph on water, what makes water cohesive.
|
Water is made cohesive due to its polar property which allows it to attract other water molecules via hydrogen bonds.
|
null | false
| null |
Why was there a storm in San Francisco last week?
|
Last week, there was a storm caused by an atmospheric river. Last Tuesday, more than 200 thousand people in the Bay area lost power because of the storm. It was one of many storms that passed through California this winter.
|
|
null | false
| null |
Richard Liu Qiangdong (Chinese: 刘强东; born March 10, 1973, or February 14, 1974[Note 1]) is a Chinese Internet entrepreneur. He has been called the "Jeff Bezos of China" and his company JD.com has been compared to Amazon due to its business model. Liu founded Jingdong as a business-to-consumer single retail store for magneto-optical products in June 1998 and later moved the company into an e-commerce website known as JD.com in 2004. Liu is the company's chief executive officer and expanded its e-commerce products from selling consumer electronics to less specialized items, which grew JD.com into one of the largest retailers in China. According to Forbes, Liu's net worth is US$9 billion as of May 2020 with a 15.8% stake in JD.com. He was included on the "China Rich List" for 2019.
|
Who is the Jeff Bezos equivalent in China, and how much of net worth is him or her?
|
Liu Qiangdong is being considered as the Jeff Bezos of China. He is the owner of JD.com, one of the largest e-shopping website in China. Liu is the CEO of the company and has 15.8% stake in it. His net worth is 9 billion US dollars as of May 2020.
|
null | false
| null |
Identify which animal species is alive or extinct: Wolly Mammoth, Poison Dart Frog
|
Poison Dart Frog is alive, Wolly Mammoth is extinct.
|
|
null | false
| null |
Classify the following water bodies into oceans, seas, rivers and lakes- aral, pacific, baikal, superior, amazon, colorado, indian, arabian, north, baltic, okhotosk, tahoe, victoria, lena, danube, arctic, mekong
|
oceans- pacific, indian, arctic
seas- okhotosk, aral, arabian, north, baltic
rivers- amazon, colorado, lena, danube, mekong
lakes- superior, tahoe, victoria , baikal
|
|
null | false
| null |
Write a brief passage or elaborate in favour of or against the use or impact of social media in our day to day life.
|
Information and communication technology has changed rapidly over the past few years. The pace of change is accelerating rapidly. Even at this stage, after a decade or so there is going to be a drastic change in whatever technology we are using around. Onto this note, along with the rise of communication technology the use of social media have risen to a different level. The main reason why people share information on social media is to reveal valuable and entertaining content to others, to grow and nourish relationships and to get the word about brands and interests they had love to use or support. All these factors have caused social media to evolve to keeping in touch with people around that could make a real impact on society. The domains of social media is being used in many ways that shape education, politics, culture, careers, innovation and many more things around. Therefore, social media has changed the way we live our lives. It has redefined the way we imagine our surroundings. Who could have imagined that community networking sites like Twitter, Instagram, Facebook and many more would become a major platform for brands to find potential customers.There are both positive and negative impacts of social media on society as well as businesses. Social media can impact you both positively and negatively. If you are a content creator, or a startup owner, then social media is a great platform for you to meet your customers and people around. However, for individuals, social media is more like an addiction which may cause discomfort if not addressed properly and may impact their mental health. Social media is a great innovation that has changed the way we communicate and interact with each other. The main benefits of social media may include staying update with current affairs going around, communicating anytime and anywhere from the comfort of the any part of the world, advertising platforms for brands, easy to build relationships and connecting with people around and an easy access to desired information, products and services. However, there are also several disadvantages onto the same like lack of understanding and thoughtfulness which could be a major challenge going ahead in coping up things around, decrease in real face to face conversations , affects social emotional connection and cyberbullying. So, we do need to be responsible while using and the awareness around is quite more important before using it freely.
|
|
null | false
| null |
If I Die in a Combat Zone, Box Me Up and Ship Me Home received warm reviews from critics. Observed by the Washington Star as possibly "the single greatest piece of work to come out of Vietnam", with equally positive reviews from The Guardian, Gloria Emerson of the New York Times and was described as a personal account of "aching clarity... A beautiful, painful book," by the New York Times Book Review. Similar reviews were given from The Times, The Washington Post, The Sunday Times, The Financial Times and Time Magazine who cited O'Brien as "Perhaps the best writer about Vietnam".
|
Extract the publications that favorably reviewed "If I Die in a Combat Zone, Box Me Up and Ship Me Home" and put them in a comma separated list.
|
The Washington Star, The Guardian, the New York Times Book Review, The Times, The Washington Post, The Sunday Times, The Financial Times, Time Magazine
|
null | false
| null |
Top 5 movies that I would love to watch over and over again
|
1. The Shawshank Redemption
2. Forrest Gump
3. Its wonderful life
4. Life Is Beautiful
5. A beautiful mind
|
|
null | false
| null |
List five titles by Anne Rice.
|
Lasher, The Queen of the Damned, The Vampire Lestat, Interview with a Vampire, Memnoch the Devil
|
|
null | false
| null |
What civilization built temples across the Yucatan peninsula?
|
The Mayan civilization was based in the Yucatan peninsula
|
|
null | false
| null |
The peanut (Arachis hypogaea), also known as the groundnut, goober, pindar or monkey nut (UK), is a legume crop grown mainly for its edible seeds. It is widely grown in the tropics and subtropics, important to both small and large commercial producers. It is classified as both a grain legume and, due to its high oil content, an oil crop. World annual production of shelled peanuts was 44 million tonnes in 2016, led by China with 38% of the world total. Atypically among legume crop plants, peanut pods develop underground (geocarpy) rather than above ground. With this characteristic in mind, the botanist Carl Linnaeus gave peanuts the specific epithet hypogaea, which means "under the earth".
The peanut belongs to the botanical family Fabaceae (or Leguminosae), commonly known as the legume, bean, or pea family. Like most other legumes, peanuts harbor symbiotic nitrogen-fixing bacteria in root nodules. The capacity to fix nitrogen means peanuts require less nitrogen-containing fertilizer and improve soil fertility, making them valuable in crop rotations.
Peanuts are similar in taste and nutritional profile to tree nuts such as walnuts and almonds, and, as a culinary nut, are often served in similar ways in Western cuisines. The botanical definition of a nut is "a fruit whose ovary wall becomes hard at maturity." Using this criterion, the peanut is not a nut. However, peanuts are usually categorized as nuts for culinary purposes and in common English more generally.
|
Based on this text, would it be correct to say a peanut is a nut?
|
This article describes different ways to classify a peanut. In a culinary or colloquial regard, yes it can be considered a nut. In the botanical or biological definition, it is not a nut but rather a legume.
|
null | false
| null |
What are the pros and cons of charging a Tesla at Public Charging Stations?
|
Public charging stations are often the fastest and most efficient ways to charge your Tesla Model Y. It can take under an hour to charge your Model Y if you use rapid chargers at a public charging point. You can find public charging stations all over the country.
Pros
Public charging stations are the only place you'll find rapid or ultra-rapid charging, although not all public charging networks provide these speeds. This is the fastest way to charge your EV. Some charging stations are free, such as Pod Point's Tesco charging points.
Cons
The biggest problem with this option is there's too much variety. Trying to figure out how to find the right charging station that meets your requirements can get complicated. Each network has different locations, different prices and different apps.
|
|
null | false
| null |
What are your 5 favourite museums in London?
|
Here are my 5 favourite museums in London:
- Natural History Museum
- The National Gallery
- British Museum
- Tate Modern
- Royal Air Force Museum
|
|
null | false
| 119
|
In his work, Caruana BIBREF4 proposed a multitask approach in which a learner takes advantage of the multiplicity of interdependent tasks while jointly learning them. The intuition is that if the tasks are correlated, the learner can learn a model jointly for them while taking into account the shared information which is expected to improve its generalization ability. People express their opinions online on various subjects (events, products..), on several languages and in several styles (tweets, paragraph-sized reviews..), and it is exactly this variety that motivates the multitask approaches. Specifically for Twitter for instance, the different settings of classification like binary, ternary and fine-grained are correlated since their difference lies in the sentiment granularity of the classes which increases while moving from binary to fine-grained problems.
There are two main decisions to be made in our approach: the learning algorithm, which learns a decision function, and the data representation. With respect to the former, neural networks are particularly suitable as one can design architectures with different properties and arbitrary complexity. Also, as training neural network usually relies on back-propagation of errors, one can have shared parts of the network trained by estimating errors on the joint tasks and others specialized for particular tasks. Concerning the data representation, it strongly depends on the data type available. For the task of sentiment classification of tweets with neural networks, distributed embeddings of words have shown great potential. Embeddings are defined as low-dimensional, dense representations of words that can be obtained in an unsupervised fashion by training on large quantities of text BIBREF10 .
Concerning the neural network architecture, we focus on Recurrent Neural Networks (RNNs) that are capable of modeling short-range and long-range dependencies like those exhibited in sequence data of arbitrary length like text. While in the traditional information retrieval paradigm such dependencies are captured using INLINEFORM0 -grams and skip-grams, RNNs learn to capture them automatically BIBREF11 . To circumvent the problems with capturing long-range dependencies and preventing gradients from vanishing, the long short-term memory network (LSTM) was proposed BIBREF12 . In this work, we use an extended version of LSTM called bidirectional LSTM (biLSTM). While standard LSTMs access information only from the past (previous words), biLSTMs capture both past and future information effectively BIBREF13 , BIBREF11 . They consist of two LSTM networks, for propagating text forward and backwards with the goal being to capture the dependencies better. Indeed, previous work on multitask learning showed the effectiveness of biLSTMs in a variety of problems: BIBREF14 tackled sequence prediction, while BIBREF6 and BIBREF15 used biLSTMs for Named Entity Recognition and dependency parsing respectively.
Figure FIGREF2 presents the architecture we use for multitask learning. In the top-left of the figure a biLSTM network (enclosed by the dashed line) is fed with embeddings INLINEFORM0 that correspond to the INLINEFORM1 words of a tokenized tweet. Notice, as discussed above, the biLSTM consists of two LSTMs that are fed with the word sequence forward and backwards. On top of the biLSTM network one (or more) hidden layers INLINEFORM2 transform its output. The output of INLINEFORM3 is led to the softmax layers for the prediction step. There are INLINEFORM4 softmax layers and each is used for one of the INLINEFORM5 tasks of the multitask setting. In tasks such as sentiment classification, additional features like membership of words in sentiment lexicons or counts of elongated/capitalized words can be used to enrich the representation of tweets before the classification step BIBREF3 . The lower part of the network illustrates how such sources of information can be incorporated to the process. A vector “Additional Features” for each tweet is transformed from the hidden layer(s) INLINEFORM6 and then is combined by concatenation with the transformed biLSTM output in the INLINEFORM7 layer.
There are two main decisions to be made in our approach: the learning algorithm, which learns a decision function, and the data representation.
|
What are the two main decisions in the author's proposed approach?
|
The learning algorithm and the data representation.
|
null | false
| null |
What is compound weightlifting?
|
Compound weightlifting is a style of weight training that uses multiple muscle groups, joints and movement patterns in order to perform the lift. Typical lifts include the deadlift, bench press, overhead press and squat.
|
|
null | false
| null |
Tell me an idea for a film based off of the cartoon movie Cars but incorporate self-driving cars.
|
A self-driving car named Walter (voiced by Will Arnett) is stuck driving drunk passengers around San Francisco at all hours of the night trying to make ends meet. He frequently has pedestrians staring at him because he doesn't have a driver, kids drawing graffiti on him, and passengers giving him a bad review for being too cautious while driving.
Walter dreams of being a race car like the rich cars he sees on Tik Tok. They get to drive fast as Sonoma Raceway and race each other in front of stadiums full of people while Walter has to follow the 30mph speed limit on SF streets.
Walter finds out that the race car life isn't all it's cracked up to be, and realizes over time to be happy with what he has and that at least he isn't one of those ride-share vehicles that can't drive themselves around town.
|
|
null | false
| null |
How do you add two or more numbers
|
You can add numbers by treating each number as a count of items and then counting all of the items to get a total result which is the sum of the numbers.
The process of adding numbers is known as addition and is one of four basic maths operations. This area of maths is known as arithmetic.
Counting the quantity of items is the same as repeatedly adding 1.
For example, adding the numbers two and three (or 2 + 3) is the same as adding one (1) two times and then adding one (1) three times.
|
|
null | false
| null |
Most avalanches occur spontaneously during storms under increased load due to snowfall and/or erosion. The second largest cause of natural avalanches is metamorphic changes in the snowpack such as melting due to solar radiation. Other natural causes include rain, earthquakes, rockfall and icefall. Artificial triggers of avalanches include skiers, snowmobiles, and controlled explosive work. Contrary to popular belief, avalanches are not triggered by loud sound; the pressure from sound is orders of magnitude too small to trigger an avalanche.
Avalanche initiation can start at a point with only a small amount of snow moving initially; this is typical of wet snow avalanches or avalanches in dry unconsolidated snow. However, if the snow has sintered into a stiff slab overlying a weak layer then fractures can propagate very rapidly, so that a large volume of snow, that may be thousands of cubic metres, can start moving almost simultaneously.
A snowpack will fail when the load exceeds the strength. The load is straightforward; it is the weight of the snow. However, the strength of the snowpack is much more difficult to determine and is extremely heterogeneous. It varies in detail with properties of the snow grains, size, density, morphology, temperature, water content; and the properties of the bonds between the grains. These properties may all metamorphose in time according to the local humidity, water vapour flux, temperature and heat flux. The top of the snowpack is also extensively influenced by incoming radiation and the local air flow. One of the aims of avalanche research is to develop and validate computer models that can describe the evolution of the seasonal snowpack over time. A complicating factor is the complex interaction of terrain and weather, which causes significant spatial and temporal variability of the depths, crystal forms, and layering of the seasonal snowpack.
|
What are the most common causes of avalanches, in descending order of likelihood (where most likely is first)?
|
The most likely causes of avalanches, in order of likelihood, are:
- Increased snow loads during storms
- Erosion in the snowpack caused by storms
- Metamorphic changes in the snowpack such as melting due to weather or rain
- Earthquakes that trigger shifting in the snowpack
- Falling objects, such as rockfall and snowfall, can trigger avalanches
- Human triggers, such as skiers, snowmobiles, and controlled explosives.
|
null | false
| null |
Explain Anthropocene to 13-year-old kid
|
The term "anthropocene" is informally used in scientific contexts. The Geological Society of America entitled its 2011 annual meeting: Archean to Anthropocene: The past is the key to the future. The new epoch has no agreed start-date, but one proposal, based on atmospheric evidence, is to fix the start with the Industrial Revolution c. 1780, with the invention of the steam engine. Other scientists link the new term to earlier events, such as the rise of agriculture and the Neolithic Revolution (around 12,000 years BP). Evidence of relative human impact – such as the growing human influence on land use, ecosystems, biodiversity, and species extinction – is substantial; scientists think that human impact has significantly changed (or halted) the growth of biodiversity. Those arguing for earlier dates posit that the proposed Anthropocene may have begun as early as 14,000–15,000 years BP, based on geologic evidence; this has led other scientists to suggest that "the onset of the Anthropocene should be extended back many thousand years";: 1 this would make the Anthropocene essentially synonymous with the current term, Holocene.
|
|
null | false
| null |
how to handle loneliness in your life?
|
The first step of handling loneliness, depression is to observe the triggering patterns and understand the causal effect. Once, we identify the time block, try to compartmentalize your feelings and back thoughts. Fill those time blocks or duration of the day with an useful activity such as engaging in a physical activity, picking up sports, cooking or try radically a new activity and immerse yourself. so that the there are no back thoughts in your mind. Plan your weekend and evenings with activities to switch off. Pick up a hobby which you are interested in. Try to go on a date or meetup to whom you can date through various forums outside of your work or a person radically in a different field.
|
|
null | false
| null |
What are some edible weeds that grow well in Maryland?
|
Many edible plants grow well in Maryland without human intervention including chickweed, purslane, dandelion, and lamb's quarters. All of these plants may volunteer in a typical suburban yard or empty lot and will grow without being watered or tended to. They are all edible and nutritious.
|
|
null | false
| 158
|
We propose an attention convolutional matching layer to integrate two similarity maps given above. That is, QACNN Layer is used to learn the location relationship pattern. It contains a two-staged CNN combined with query-based attention mechanism. Each stage comprises two major part: attention map and output representation.
Fig. FIGREF10 shows the architecture of the attention map in first stage CNN. We choose INLINEFORM0 sentence slice INLINEFORM1 in INLINEFORM2 , and apply CNN to it using the convolution kernel INLINEFORM3 , where superscript INLINEFORM4 denotes attention map, subscript INLINEFORM5 denotes the first stage CNN. Symbol INLINEFORM6 and INLINEFORM7 represent width of kernel and number of kernel respectively. The generated feature INLINEFORM8 is as follow: DISPLAYFORM0
where INLINEFORM0 is the bias. With INLINEFORM1 covering whole query and several words in the passage, convolution kernels would learn the query syntactic structure and give weight to each passage's location. That's why we use sigmoid function as activation function in this stage. Furthermore, we perform maxpooling to INLINEFORM2 perpendicularly in order to find the largest weight between different kernels in the same location, using maxpool kernel shaped INLINEFORM3 , and then generate word-level attention map INLINEFORM4 for each sentence.
In this stage, we want to acquire passage's sentence features based on the query and the choice respectively. We apply CNN to INLINEFORM0 to aggregate pattern of location relationship and acquire choice-based sentence features. Also, we apply CNN to INLINEFORM1 to acquire query-based sentence features. CNN architecture of output representation part Fig. FIGREF11 is similar to which of attention map part, but we use different kernels INLINEFORM2 and different bias INLINEFORM3 : DISPLAYFORM0
where the superscript INLINEFORM0 denotes output representation. We apply INLINEFORM3 on INLINEFORM4 and INLINEFORM5 then finally generate INLINEFORM6 and INLINEFORM7 using eq.4. We then multiply INLINEFORM8 by the word-level attention map INLINEFORM9 which comes from stage 2.2.1 element-wise through the first dimension. At last, we maxpool INLINEFORM10 and INLINEFORM11 horizontally with kernel shape INLINEFORM12 to get the query-based sentence features INLINEFORM13 and choice-based sentence features INLINEFORM14 .
Fig. FIGREF28 is the architecture of attention map in the second stage CNN. Based on first stage query-based sentence features from section UID15 , we want to acquire sentence-level attention map. The input of this stage is INLINEFORM0 , which will be further refined by CNN with kernel INLINEFORM1 and generates intermediate features INLINEFORM2 . DISPLAYFORM0
Then, same as attention map of first stage, we maxpool INLINEFORM0 with kernel shaped INLINEFORM1 , and obtain sentence-level attention map INLINEFORM2 .
Output representation part of the second stage in Fig. FIGREF29 has two input, sentence-level attention map INLINEFORM0 and sentence-level features INLINEFORM1 . The equations here are similar to those previously mentioned. As follow: DISPLAYFORM0
where INLINEFORM0 , INLINEFORM1 , and INLINEFORM2 . Output representation of certain choice INLINEFORM3 is the final output of QACNN Layer.
Output Representation of First Stage In this stage, we want to acquire passage’s sentence features based on the query and the choice respectively.
|
What will be acquired in the output representation of the first stage?
|
Passage' s sentence features based on the query and the choice respectively.
|
null | false
| null |
Qumran (Hebrew: קומראן; Arabic: خربة قمران Khirbet Qumran) is an archaeological site in the West Bank managed by Israel's Qumran National Park. It is located on a dry marl plateau about 1.5 km (1 mi) from the northwestern shore of the Dead Sea, near the Israeli settlement and kibbutz of Kalya.
The Hellenistic period settlement was constructed during the reign of Hasmonean leader John Hyrcanus (134–104 BCE) or somewhat later. Qumran was inhabited by a Jewish community of the late Second Temple period, which most scholars identify with the mystic sect of the Essenes; however, other groups were also suggested. It was occupied most of the time until 68 CE and was destroyed by the Romans during the First Jewish-Roman War, possibly as late as 73 CE. It was later used by Jewish rebels during the Bar Kokhba Revolt. Today, the Qumran site is best known as the settlement nearest to the Qumran Caves where the Dead Sea Scrolls were hidden, caves in the sheer desert cliffs and beneath, in the marl terrace. The principal excavations at Qumran were conducted by Roland de Vaux in the 1950s, and several later digs have been carried out.
Since the 1967 Six Day War, Qumran has been managed by the Israel Nature and Parks Authority.
|
Is Qumran worth to visit
|
Qumran is an archaeological site in the West Bank managed by Israel's Qumran National Park. It is located on a dry marl plateau about 1.5 km (1 mi) from the northwestern shore of the Dead Sea, near the Israeli settlement and kibbutz of Kalya.
The Hellenistic period settlement was constructed during the reign of Hasmonean leader John Hyrcanus (134–104 BCE) or somewhat later. Qumran was inhabited by a Jewish community of the late Second Temple period, which most scholars identify with the mystic sect of the Essenes; however, other groups were also suggested. It was occupied most of the time until 68 CE and was destroyed by the Romans during the First Jewish-Roman War, possibly as late as 73 CE. It was later used by Jewish rebels during the Bar Kokhba Revolt. Today, the Qumran site is best known as the settlement nearest to the Qumran Caves where the Dead Sea Scrolls were hidden, caves in the sheer desert cliffs and beneath, in the marl terrace. The principal excavations at Qumran were conducted by Roland de Vaux in the 1950s, and several later digs have been carried out.
Since the 1967 Six Day War, Qumran has been managed by the Israel Nature and Parks Authority.
|
null | false
| null |
What should I make for dinner tonight?
|
-green curry with all the vegetables in your fridge right now.
-pasta bolognese.
-white beans and dark leafy greens stew.
|
|
null | false
| null |
How many provinces contain the word ‘Holland’?
|
There are 6 provinces in the Netherlands that contain the word ‘Holland’. These provinces are North Holland, South Holland, Zeeland, Utrecht, Flevoland, and Gelderland.
|
|
null | false
| null |
Which one is best for youtube creator gorpro or DJI Osmo pocket 2
|
OSMO pocket 2 is best as it has gimble, if you are a creator you will go for day to day videography , OSMO will not disappoint you
|
|
null | false
| null |
What is an audible?
|
An audible can be used to describe how one can be prepared for changing circumstances. When asked an unexpected question, one is able to pivot their prepared response for another.
|
|
null | false
| 183
|
We automatically create a test set from the OpenSubtitles corpus BIBREF22 . The goal is to provide a large number of difficult test cases where an English pronoun has to be translated to a German pronoun.
The most challenging cases are translating it to either er, sie or es, depending on the grammatical gender of the antecedent. Not only is the translation of it ambiguous, there is also class imbalance in the training data (see Table TABREF18 ). There is roughly a 30% probability that it is aligned to es, which makes it difficult to learn to translate er and sie. We use parsing and automatic co-reference resolution to find translation pairs that satisfy our constraints.
To provide a basis for filtering with constraints, we tokenize the whole data set with the Moses tokenizer, generate symmetric word alignments with fast_align BIBREF23 , parse the English text with CoreNLP BIBREF24 , parse the German text with ParZu BIBREF25 and perform coreference resolution on both sides. The coreference chains are obtained with the neural model of CoreNLP for English, and with CorZu for German BIBREF26 , respectively.
Then we opt for high-precision, aggressive filtering, according to the following protocol: for each pair of sentences INLINEFORM0 in English and German, extract iff
INLINEFORM0 contains the English pronoun it, and INLINEFORM1 contains a German pronoun that is third person singular (er, sie or es), as indicated by their part-of-speech tags;
those pronouns are aligned to each other;
both pronouns are in a coreference chain;
their nominal antecedents in the coreference chain are aligned on word level.
This removes most candidate pairs, but is necessary to overcome the noise introduced by our preprocessing pipeline, most notably coreference resolution. From the filtered set, we create a balanced test set by randomly sampling 4000 instances of each of the three translations of it under consideration (er, sie, es). We do not balance antecedent distance. See Table TABREF25 for the distribution of pronoun pairs and antecedent distance in the test set.
For each sentence pair in the resulting test set, we introduce contrastive translations. A contrastive translation is a translation variant where the correct pronoun is swapped with an incorrect one. For an example, see Table TABREF19 , where the pronoun it in the original translation corresponds to sie because the antecedent bat is a feminine noun in German (Fledermaus). We produce wrong translations by replacing sie with one of the other pronouns (er, es).
Note that, by themselves, these contrastive translations are grammatically correct if the antecedent is outside the current sentence. The test set also contains pronouns with an antecedent in the same sentence (antecedent distance 0). Those examples do not require any additional context for disambiguation and we therefore expect the sentence-level baseline to perform well on them.
We take extra care to ensure that the resulting contrastive translations are grammatically correct, because ungrammatical sentences are easily dismissed by an NMT system. For instance, if there are any possessive pronouns (such as seine) in the sentence, we also change their gender to match the personal pronoun replacement.
The German coreference resolution system does not resolve es because most instances of es in German are either non-referential forms, or they refer to a clause instead of a nominal antecedent. We limit the test set to nominal antecedents, as these are the only ambiguous cases with respect to translation. For this reason, we have to rely entirely on the English coreference links for the extraction of sentence pairs with it INLINEFORM0 es, as opposed to pairs with it INLINEFORM1 er and it INLINEFORM2 sie where we have coreference chains in both languages.
Our extraction process respects document boundaries, to ensure we always search for the right context. We extract additional information from the annotated documents, such as the distance (in sentences) between pronouns and their antecedents, the document of origin, lemma, morphology and dependency information if available.
A contrastive translation is a translation variant where the correct pronoun is swapped with an incorrect one.
|
What is a contrastive translation?
|
A contrastive translation is a translation variant where the correct pronoun is swapped with an incorrect one.
|
null | false
| null |
Who was Christopher Columbus?
|
Christopher Columbus was an Italian explorer and navigator who completed four voyages across the Atlantic ocean. Scholars believe he was born between August 25th and October 31st 1451 and he is know for the expeditions that inaugurated a period of exploration in Europe that resulted in the colonization of many countries in the Americas that lasted for centuries. Many places in the western hemisphere bear his name including Colombia (the country), British Columbia (province in Canada), District of Columbia (capital of the US) and Columbus (city in Ohio). Christopher Columbus died on May 20th, 1506.
|
|
1908.11546
| false
| null |
FLOAT SELECTED: Table 5: Entity F1 and Success F1 at dialogue level.
FLOAT SELECTED: Table 5: Entity F1 and Success F1 at dialogue level.
|
How better is gCAS approach compared to other approaches?
|
For entity F1 in the movie, taxi and restaurant domain it results in scores of 50.86, 64, and 60.35. For success, it results it outperforms in the movie and restaurant domain with scores of 77.95 and 71.52
|
null | false
| null |
There have been twenty permanent and eight caretaker managers of Arsenal since 1897; Stewart Houston has managed the club in two separate spells as caretaker. The most successful person to manage Arsenal is Arsène Wenger, who won three Premier League titles, seven FA Cups and seven Community Shields between 1996 and 2018. Wenger is the club's longest-serving manager; he surpassed George Allison's record of 13 years in October 2009. Two Arsenal managers have died in the job – Herbert Chapman and Tom Whittaker.
|
How many managers have Arsenal had?
|
Arsenal have had twenty permanent and eight caretake managers.
|
null | false
| null |
Classify each of the following as either a city or a country: Delhi, China, Tokyo, Australia, India, Kuala Lumpur, Egypt, Morocco, Paris, London, Finland, Dubai, Angola, Chennai, Cape Town, Guinea, Ivory Coast, Ecuador, Sao Paulo, Toronto, Iran, Moscow, Berlin, Slovenia, Prague, Italy, Milan, Argentina, Cuba, Cambodia, Brunei, Jakarta, Manila
|
The following are countries - China, Australia, India, Egypt, Morocco, Finland, Angola, Guinea, Ivory Coast, Ecuador, Iran, Slovenia, Italy, Argentina, Cuba, and the Philippines
The following are cities - Delhi, Tokyo, Kuala Lumpur, Paris, London, Dubai, Chennai, Cape Town, Sao Paulo, Toronto, Moscow, Berlin, Prague, Milan, Jakarta, and Manila
|
|
null | false
| null |
In 1871, at the age of eight, Narendranath enrolled at Ishwar Chandra Vidyasagar's Metropolitan Institution, where he went to school until his family moved to Raipur in 1877. In 1879, after his family's return to Calcutta, he was the only student to receive first-division marks in the Presidency College entrance examination. He was an avid reader in a wide range of subjects, including philosophy, religion, history, social science, art and literature. He was also interested in Hindu scriptures, including the Vedas, the Upanishads, the Bhagavad Gita, the Ramayana, the Mahabharata and the Puranas. Narendra was trained in Indian classical music, and regularly participated in physical exercise, sports and organised activities. Narendra studied Western logic, Western philosophy and European history at the General Assembly's Institution (now known as the Scottish Church College). In 1881, he passed the Fine Arts examination, and completed a Bachelor of Arts degree in 1884. Narendra studied the works of David Hume, Immanuel Kant, Johann Gottlieb Fichte, Baruch Spinoza, Georg W. F. Hegel, Arthur Schopenhauer, Auguste Comte, John Stuart Mill and Charles Darwin. He became fascinated with the evolutionism of Herbert Spencer and corresponded with him, translating Herbert Spencer's book Education (1861) into Bengali. While studying Western philosophers, he also learned Sanskrit scriptures and Bengali literature.
|
As per the passage whose works did Narendranath study?
|
David Hume, Immanuel Kant, Johann Gottlieb Fichte, Baruch Spinoza, Georg W. F. Hegel, Arthur Schopenhauer, Auguste Comte, John Stuart Mill and Charles Darwin.
|
null | false
| null |
What is the difference between mozzarella and cheddar?
|
Mozzarella (or Mozz) can be served fresh, while cheddar is generally aged for some time to improve flavor before it is consumed.
|
|
null | false
| 4
|
Knowledge Base Question Answering (KBQA) systems answer questions by obtaining information from KB tuples BIBREF0 , BIBREF1 , BIBREF2 , BIBREF3 , BIBREF4 , BIBREF5 . For an input question, these systems typically generate a KB query, which can be executed to retrieve the answers from a KB. Figure 1 illustrates the process used to parse two sample questions in a KBQA system: (a) a single-relation question, which can be answered with a single $<$ head-entity, relation, tail-entity $>$ KB tuple BIBREF6 , BIBREF7 , BIBREF2 ; and (b) a more complex case, where some constraints need to be handled for multiple entities in the question. The KBQA system in the figure performs two key tasks: (1) entity linking, which links $n$ -grams in questions to KB entities, and (2) relation detection, which identifies the KB relation(s) a question refers to.
The main focus of this work is to improve the relation detection subtask and further explore how it can contribute to the KBQA system. Although general relation detection methods are well studied in the NLP community, such studies usually do not take the end task of KBQA into consideration. As a result, there is a significant gap between general relation detection studies and KB-specific relation detection. First, in most general relation detection tasks, the number of target relations is limited, normally smaller than 100. In contrast, in KBQA even a small KB, like Freebase2M BIBREF2 , contains more than 6,000 relation types. Second, relation detection for KBQA often becomes a zero-shot learning task, since some test instances may have unseen relations in the training data. For example, the SimpleQuestions BIBREF2 data set has 14% of the golden test relations not observed in golden training tuples. Third, as shown in Figure 1 (b), for some KBQA tasks like WebQuestions BIBREF0 , we need to predict a chain of relations instead of a single relation. This increases the number of target relation types and the sizes of candidate relation pools, further increasing the difficulty of KB relation detection. Owing to these reasons, KB relation detection is significantly more challenging compared to general relation detection tasks.
This paper improves KB relation detection to cope with the problems mentioned above. First, in order to deal with the unseen relations, we propose to break the relation names into word sequences for question-relation matching. Second, noticing that original relation names can sometimes help to match longer question contexts, we propose to build both relation-level and word-level relation representations. Third, we use deep bidirectional LSTMs (BiLSTMs) to learn different levels of question representations in order to match the different levels of relation information. Finally, we propose a residual learning method for sequence matching, which makes the model training easier and results in more abstract (deeper) question representations, thus improves hierarchical matching.
In order to assess how the proposed improved relation detection could benefit the KBQA end task, we also propose a simple KBQA implementation composed of two-step relation detection. Given an input question and a set of candidate entities retrieved by an entity linker based on the question, our proposed relation detection model plays a key role in the KBQA process: (1) Re-ranking the entity candidates according to whether they connect to high confident relations detected from the raw question text by the relation detection model. This step is important to deal with the ambiguities normally present in entity linking results. (2) Finding the core relation (chains) for each topic entity selection from a much smaller candidate entity set after re-ranking. The above steps are followed by an optional constraint detection step, when the question cannot be answered by single relations (e.g., multiple entities in the question). Finally the highest scored query from the above steps is used to query the KB for answers.
Our main contributions include: (i) An improved relation detection model by hierarchical matching between questions and relations with residual learning; (ii) We demonstrate that the improved relation detector enables our simple KBQA system to achieve state-of-the-art results on both single-relation and multi-relation KBQA tasks.
First, in order to deal with the unseen relations, we propose to break the relation names into word sequences for question-relation matching.
|
How did they deal with the unseen relations in the improved approach？
|
We proposed to break the relation names into word sequences for question-relation matching.
|
null | false
| 105
|
Since humans amass more and more generally available data in the form of unstructured text it would be very useful to teach machines to read and comprehend such data and then use this understanding to answer our questions. A significant amount of research has recently focused on answering one particular kind of questions the answer to which depends on understanding a context document. These are cloze-style questions BIBREF0 which require the reader to fill in a missing word in a sentence. An important advantage of such questions is that they can be generated automatically from a suitable text corpus which allows us to produce a practically unlimited amount of them. That opens the task to notoriously data-hungry deep-learning techniques which now seem to outperform all alternative approaches.
Two such large-scale datasets have recently been proposed by researchers from Google DeepMind and Facebook AI: the CNN/Daily Mail dataset BIBREF1 and the Children's Book Test (CBT) BIBREF2 respectively. These have attracted a lot of attention from the research community BIBREF3 , BIBREF4 , BIBREF5 , BIBREF6 , BIBREF7 , BIBREF8 , BIBREF9 , BIBREF10 , BIBREF11 , BIBREF12 with a new state-of-the-art model coming out every few weeks.
However if our goal is a production-level system actually capable of helping humans, we want the model to use all available resources as efficiently as possible. Given that
we believe that if the community is striving to bring the performance as far as possible, it should move its work to larger data.
This thinking goes in line with recent developments in the area of language modelling. For a long time models were being compared on several "standard" datasets with publications often presenting minuscule improvements in performance. Then the large-scale One Billion Word corpus dataset appeared BIBREF15 and it allowed Jozefowicz et al. to train much larger LSTM models BIBREF16 that almost halved the state-of-the-art perplexity on this dataset.
We think it is time to make a similar step in the area of text comprehension. Hence we are introducing the BookTest, a new dataset very similar to the Children's Book test but more than 60 times larger to enable training larger models even in the domain of text comprehension. Furthermore the methodology used to create our data can later be used to create even larger datasets when the need arises thanks to further technological progress.
We show that if we evaluate a model trained on the new dataset on the now standard Children's Book Test dataset, we see an improvement in accuracy much larger than other research groups achieved by enhancing the model architecture itself (while still using the original CBT training data). By training on the new dataset, we reduce the prediction error by almost one third. On the named-entity version of CBT this brings the ensemble of our models to the level of human baseline as reported by Facebook BIBREF2 . However in the final section we show in our own human study that there is still room for improvement on the CBT beyond the performance of our model.
Furthermore the methodology used to create our data can later be used to create even larger datasets when the need arises thanks to further technological progress.
|
Whether the methodology used to create the data can later be used to create even larger datasets when the need arises?
|
Yes.
|
null | false
| 78
|
In this paper, we proposed a model that utilizes chunked timelines of tweets and a recurrent neural model in order to infer the factuality of a Twitter news account. Our experimental results indicate the importance of analyzing tweet stream into chunks, as well as the benefits of heterogeneous knowledge source (i.e., lexica as well as text) in order to capture factuality. In future work, we would like to extend this line of research with further in-depth analysis to understand the flow change of the used features in the accounts' streams. Moreover, we would like to take our approach one step further incorporating explicit temporal information, e.g., using timestamps. Crucially, we are also interested in developing a multilingual version of our approach, for instance by leveraging the now ubiquitous cross-lingual embeddings BIBREF22, BIBREF23.
In future work, we would like to extend this line of research with further in-depth analysis to understand the flow change of the used features in the accounts' streams. Moreover, we would like to take our approach one step further incorporating explicit temporal information, e.g., using timestamps. Crucially, we are also interested in developing a multilingual version of our approach, for instance by leveraging the now ubiquitous cross-lingual embeddings.
|
What are the author's plans for his future work?
|
To extend this line of research with further in-depth analysis to understand the flow change of the used features in the accounts' streams, take their approach one step further incorporating explicit temporal information, e.g., using timestamps, and develop a multilingual version of their approach, for instance by leveraging the now ubiquitous cross-lingual embeddings.
|
null | false
| null |
Tell me whether each of these countries is north or south of the equator: Finland, Canada, South Africa, Iceland, Germany, Uruguay, Zimbabwe
|
Finland: North
Canada: North
South Africa: South
Iceland: North
Germany: North
Uruguay: South
Zimbabwe: South
|
|
null | false
| null |
Why does my dog dig in the furniture?
|
Most dogs were originally bred to service a function. Some dogs were bred to chase small animals, for hunting rabbits or fox. Many of these breeds were also intended to dig for their kills, so your dog may burrow because it's born to do that. Said differently, some terrier breeds were meant to hunt rats, i.e. the rat terrier. These dogs will dig even if there is nothing to dig for simply because it's what they were bred to do. Other dogs dig to create nesting areas, to warm up their area before they lay down, or to check for bugs.
|
|
null | false
| 57
|
The vast amounts of data collected by healthcare providers in conjunction with modern data analytics techniques present a unique opportunity to improve health service provision and the quality and safety of medical care for patient benefit BIBREF0 . Much of the recent research in this area has been on personalised medicine and its aim to deliver better diagnostics aided by the integration of diverse datasets providing complementary information. Another large source of healthcare data is organisational. In the United Kingdom, the National Health Service (NHS) has a long history of documenting extensively the different aspects of healthcare provision. The NHS is currently in the process of increasing the availability of several databases, properly anonymised, with the aim of leveraging advanced analytics to identify areas of improvement in NHS services.
One such database is the National Reporting and Learning System (NRLS), a central repository of patient safety incident reports from the NHS in England and Wales. Set up in 2003, the NRLS now contains more than 13 million detailed records. The incidents are reported using a set of standardised categories and contain a wealth of organisational and spatio-temporal information (structured data), as well as, crucially, a substantial component of free text (unstructured data) where incidents are described in the `voice' of the person reporting. The incidents are wide ranging: from patient accidents to lost forms or referrals; from delays in admission and discharge to serious untoward incidents, such as retained foreign objects after operations. The review and analysis of such data provides critical insight into the complex functioning of different processes and procedures in healthcare towards service improvement for safer carer.
Although statistical analyses are routinely performed on the structured component of the data (dates, locations, assigned categories, etc), the free text remains largely unused in systematic processes. Free text is usually read manually but this is time-consuming, meaning that it is often ignored in practice, unless a detailed review of a case is undertaken because of the severity of harm that resulted. There is a lack of methodologies that can summarise content and provide content-based groupings across the large volume of reports submitted nationally for organisational learning. Methods that could provide automatic categorisation of incidents from the free text would sidestep problems such as difficulties in assigning an incident category by virtue of a priori pre-defined lists in the reporting system or human error, as well as offering a unique insight into the root cause analysis of incidents that could improve the safety and quality of care and efficiency of healthcare services.
Our goal in this work is to showcase an algorithmic methodology that detects content-based groups of records in a given dataset in an unsupervised manner, based only on the free and unstructured textual description of the incidents. To do so, we combine recently developed deep neural-network high-dimensional text-embedding algorithms with network-theoretical methods. In particular, we apply multiscale Markov Stability (MS) community detection to a sparsified geometric similarity graph of documents obtained from text vector similarities. Our method departs from traditional natural language processing tools, which have generally used bag-of-words (BoW) representation of documents and statistical methods based on Latent Dirichlet Allocation (LDA) to cluster documents BIBREF1 . More recent approaches have used deep neural network based language models clustered with k-means, without a full multiscale graph analysis BIBREF2 . There have been some previous applications of network theory to text analysis. For example, Lanchichinetti and co-workers BIBREF3 used a probabilistic graph construction analysed with the InfoMap algorithm BIBREF4 ; however, their community detection was carried out at a single-scale and the representation of text as BoW arrays lacks the power of neural network text embeddings. The application of multiscale community detection allows us to find groups of records with consistent content at different levels of resolution; hence the content categories emerge from the textual data, rather than fitting with pre-designed classifications. The obtained results could thus help mitigate possible human error or effort in finding the right category in complex category classification trees.
We showcase the methodology through the analysis of a dataset of patient incidents reported to the NRLS. First, we use the 13 million records collected by the NRLS since 2004 to train our text embedding (although a much smaller corpus can be used). We then analyse a subset of 3229 records reported from St Mary's Hospital, London (Imperial College Healthcare NHS Trust) over three months in 2014 to extract clusters of incidents at different levels of resolution in terms of content. Our method reveals multiple levels of intrinsic structure in the topics of the dataset, as shown by the extraction of relevant word descriptors from the grouped records and a high level of topic coherence. Originally, the records had been manually coded by the operator upon reporting with up to 170 features per case, including a two-level manual classification of the incidents. Therefore, we also carried out an a posteriori comparison against the hand-coded categories assigned by the reporter (healthcare personnel) at the time of the report submission. Our results show good overall correspondence with the hand-coded categories across resolutions and, specifically, at the medium level of granularity. Several of our clusters of content correspond strongly to well-defined categories, yet our results also reveal complementary categories of incidents not defined in the external classification. In addition, the tuning of the granularity afforded by the method can be used to provide a distinct level of resolution in certain areas corresponding to specialise or particular sub-themes.
In particular, we apply multiscale Markov Stability (MS) community detection to a sparsified geometric similarity graph of documents obtained from text vector similarities.
|
What does multiscale Markov Stability (MS) community detection apply to?
|
Apply to a sparsified geometric similarity graph of documents obtained from text vector similarities.
|
null | false
| null |
Le Souvenir français is a French association for maintaining war memorials and war memory, comparable to the Commonwealth War Graves Commission.
Médaille du Souvenir Français.gif
It was created in 1887 in the departments of Alsace and Lorraine. Young women in traditional dress had been furtively placing cockades on soldiers' tombs. An Alsatian professor, Xavier Niessen, against Prussian orders in these departments, was keen to show his membership of the French fatherland and thought that remembering those who had died for France allowed the feeling of national unity to be maintained. Thus, on 7 March 1888 he summoned the French to join his new association, which then had a highly active period.
It is one of France's oldest associations d’utilité publique (being recognised as such on 1 February 1906) and has three aims:
to conserve the memory of those who have died for France
to maintain memorials to France's war dead
to hand down the memory of them to future generations
Many of its supporters were mobilised in the First World War, with the remaining ones looking after war graves and accompanying the families of the war dead on visits to them. The association was then in charge of 88,000 burials from the Franco-Prussian War and could only handle 1,700,000 burials of dead from the First World War. Thus the law of 31 July 1920 created the "service national des sépultures", took charge of military cemeteries and organised nationals necropolises. After the Second World War general Lacapelle, president of Le Souvenir Français, demanded the identification of 100,000 French soldiers who had died the battle of France and the maintenance of their tombs.
It has a general delegation by department, around 1,450 local committees in France and 52 foreign delegations. It now has 130,000 active members.
|
Please provide a brief summary of Le Souvenir français.
|
Le Souvenir français is a French association for maintaining war memorials and war memory, created in 1887 in Alsace and Lorraine. It has three aims: to conserve the memory of those who have died for France, maintain memorials to France's war dead, and hand down the memory of them to future generations. It has a general delegation by department, 1,450 local committees in France and 52 foreign delegations, and now has 130,000 active members.
|
null | false
| null |
The 522-million-worker Indian labour force is the world's second-largest, as of 2017. The service sector makes up 55.6% of GDP, the industrial sector 26.3% and the agricultural sector 18.1%. India's foreign exchange remittances of US$100 billion in 2022, highest in the world, were contributed to its economy by 32 million Indians working in foreign countries. Major agricultural products include: rice, wheat, oilseed, cotton, jute, tea, sugarcane, and potatoes.
|
From the given paragraph, extract the industry sector and the corresponding percentage of GDP in the format {Industry sector} - {Percentage} and list them using bullets
|
Here are the industries mentioned in the text and their GDP percentages:
Service sector - 55.6%
Industrial sector - 26.3%
Agricultural sector - 18.1%
|
null | false
| null |
Why Pouilly Fumé wine going well with pesto spaghetti?
|
Loire valley wine goes really well with herbal / herbaceous sauce. Loire valley Wine typically has high acidity, with balance smokiness as well as lovely mineral "tension" so it generally goes really well with the pasta sauce like pesto. Pouilly Fumé is exceptional right bank example, and there are some producers in meneton-salon making really good same style while more affordable.
|
|
null | false
| null |
Who is the author of Harry Potter?
|
J.K. Rowling is the British author of the seven-volume Harry Potter series
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.