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
| null |
How to tell if someone told a good joke?
|
You can tell if someone said a good joke if it makes a person laugh or giggle after it was communicated. You can also tell if a good joke was said by an audience reaction to the joke, if they cheer, or if they boo. Booing is a sign the joke was not so good, and cheering is a sign the joke was good. A reaction of silence to a joke means the joke was not understood. If you or an audience don't understand the context of joke make sure to ask for clarification so you can take part in cheering or booing depending on how you respond to it.
|
|
null | false
| null |
SM U-30[Note 1] was one of 329 U-boat submarines serving in the Imperial German Navy in World War I. She engaged in commerce warfare as part of the First Battle of the Atlantic. U-30 is significant for the torpedoing of the US tanker Gulflight on 1 May 1915 20 nautical miles (37 km; 23 mi) west of Scilly.
|
What kind of u-boat was the SM U-30
|
SM U-30 was one of 329 U-boat submarines serving in the Imperial German Navy in World War I. She engaged in commerce warfare as part of the First Battle of the Atlantic. U-30 is significant for the torpedoing of the US tanker Gulflight on 1 May 1915 20 nautical miles (37 km; 23 mi) west of Scilly.
|
null | false
| null |
Who saved Andromeda from the sea monster
|
Perseus
|
|
null | false
| 343
|
The use of categorical attributes (e.g., user, topic, aspects) in the sentiment analysis community BIBREF0, BIBREF1, BIBREF2 is widespread. Prior to the deep learning era, these information were used as effective categorical features BIBREF3, BIBREF4, BIBREF5, BIBREF6 for the machine learning model. Recent work has used them to improve the overall performance BIBREF7, BIBREF8, interpretability BIBREF9, BIBREF10, and personalization BIBREF11 of neural network models in different tasks such as sentiment classification BIBREF12, review summarization BIBREF13, and text generation BIBREF8.
In particular, user and product information have been widely incorporated in sentiment classification models, especially since they are important metadata attributes found in review websites. BIBREF12 first showed significant accuracy increase of neural models when these information are used. Currently, the accepted standard method is to use them as additional biases when computing the weights $a$ in the attention mechanism, as introduced by BIBREF7 as:
where $u$ and $p$ are the user and product embeddings, and $h$ is a word encoding from BiLSTM. Since then, most of the subsequent work attempted to improve the model by extending the model architecture to be able to utilize external features BIBREF14, handle cold-start entities BIBREF9, and represent user and product separately BIBREF15.
Intuitively, however, this method is not the ideal method to represent and inject attributes because of two reasons. First, representing attributes as additional biases cannot model the relationship between the text and attributes. Rather, it only adds a user- and product-specific biases that are independent from the text when calculating the attention weights. Second, injecting the attributes in the attention mechanism means that user and product information are only used to customize how the model choose which words to focus on, as also shown empirically in previous work BIBREF7, BIBREF15. However, we argue that there are more intuitive locations to inject the attributes such as when contextualizing words to modify their sentiment intensity.
We propose to represent user and product information as weight matrices (i.e., $W$ in the equation above). Directly incorporating these attributes into $W$ leads to large increase in parameters and subsequently makes the model difficult to optimize. To mitigate these problems, we introduce chunk-wise importance weight matrices, which (1) uses a weight matrix smaller than $W$ by a chunk size factor, and (2) transforms these matrix into gates such that it corresponds to the relative importance of each neuron in $W$. We investigate the use of this method when injected to several locations in the base model: word embeddings, BiLSTM encoder, attention mechanism, and logistic classifier.
The results of our experiments can be summarized in three statements. First, our preliminary experiments show that doing bias-based attribute representation and attention-based injection is not an effective method to incorporate user and product information in sentiment classification models. Second, despite using only a simple BiLSTM with attention classifier, we significantly outperform previous state-of-the-art models that use more complicated architectures (e.g., models that use hierarchical models, external memory networks, etc.). Finally, we show that these attribute representations transfer well to other tasks such as product category classification and review headline generation.
Second, despite using only a simple BiLSTM with attention classifier, we significantly outperform previous state-of-the-art models that use more complicated architectures (e.g., models that use hierarchical models, external memory networks, etc.).
|
Do the authors' models outperform previous state-of-the-art models that use more complicated architectures?
|
Yes, they do.
|
null | false
| 509
|
In our experimental setup, an ideal context encoder should capture the generalizable information for task inference, namely the difference between reward/dynamics functions across a distribution of tasks. However, as discussed in Section 1, there are two major challenges that impede conventional COMRL algorithms from learning robust representations: MDP ambiguity arises due to COMRL algorithms' sensitivity to fixed dataset distributions. Take Sparse-Point-Robot for example, as in Figure, for tasks with a goal on the semicircle, the state-action distribution exhibits specific pattern which may reflect task identity. Given D = {(s, a, s , r)} as input, the context encoder may learn a spurious correlation between state-action distributions and task identity, which causes performance degradation under distribution shifts (Table).
Sparse reward in meta-environments could exacerbate MDP ambiguity by making a considerable portion of transitions uninformative for task inference, such as the samples outside any goals in Figure. Attention mechanism, especially the batch-wise channel attention, helps the context encoder attend to the informative portion of the input transitions, and therefore significantly improve the robustness of the learned policies.
To demonstrate the robustness of FOCAL++ in presence of the two challenges above, we tested it against distribution shift by using datasets of various qualities: expert, medium, random and mixed which combines all three. Shown in Table, we observe that overall the performance drop due to distribution shift is significantly lower when attention and contrastive learning are applied.
Moreover, we are aware that even mixing of datasets generated by different behavior policies cannot fully eliminate the risk of MDP ambiguity since the state-action distributions for each task still do not completely overlap. To show that the attention modules introduced by FOCAL++ indeed works as intended by capturing the reward-task dependency, we create a new dataset on Sparse-Point-Robot by merging the state-action support across all tasks and relabelling the sparse reward according to the task-specific reward functions. In principle, this fully prevents information leakage from the state-action distributions, forcing the context encoder to learn to distinguish the reward functions between tasks while minimizing the contrastive loss. Shown in Figure, we experimented with 3 attention variants of FOCAL++ on the relabeled dataset, and found that batch-wise attention significantly improves the performance as intended. Additionally, we visualize the density distribution of batch-wise attention weights assigned to samples in Figure. We see a clear tendency for the batch-attention module to assign zero weight to samples with zero rewards (the absolutely sparse data points which lie outside all goal circles in Figure) and maximum weights to the non-zero-reward transitions, with binary classification AUC = 0.969, which is clear evidence of FOCAL++ learning the correct correlation for task inference by attending to the informative context.
We evaluate FOCAL++ on 6 continuous control meta-environments of robotic locomotion (Todorov et al., 2012) adopted from FOCAL. 4 (Sparse-Point-Robot, Sparse-Cheetah-Vel, Sparse-Cheetah-FwdBack, Sparse-Ant-Fwd-Back) and 2 (Point-Robot-Wind, Walker-2D-Params) environments require adaptation by reward and transition functions respectively. For inference, FOCAL++ aggregates context from a fixed test set to infer task embedding, and is subsequently evaluated online. Besides FOCAL, three other baselines are compared: an offline variant of the PEARL algorithm (Rakelly et al., 2019) (Batch PEARL), a context-based offline BCQ algorithm (Fujimoto et al., 2019) (Contextual BCQ) and a two-stage COMRL algorithm with reward/dynamics relabelling (Li et al., 2019) (MBML).****Figure 5: Result on the relabeled Sparse-Point-Robot dataset. (a) State distributions of the expert datasets for 20 distinct tasks, with goals uniformly distributed on a semicircle. (b) On mixed dataset, FOCAL completely fails in this scenario whereas FOCAL++ variants with batch-wise attention are able to learn. (c) Probability distribution of the batch-wise attention weight of samples with absolutely zero and non-zero reward. Binary classification AUC = 0.969.
|
The approach was evaluated on only one dataset. It is difficult to draw conclusions on the generality of the approach, right?
|
We thoroughly tested FOCAL++ on 6 MuJoCo datasets with sparse rewards, all of which are common meta-RL benchmarks and aligned with well-known papers like PEARL [1] and MQL [2], with one additional relabeled dataset for discussion of MDP ambiguity (Figure 5). As some reviewers suggested, we also added more experiments with other COMRL baselines on non-sparse environments. Please check our General Response 2. Overall, we are really grateful for your comments and advice, which are all valuable in helping us improve the work. Hope our explanation and updated material address your concerns. Please let us know if you have any further comments or concerns or ways in which we can further improve our paper.Our proposed method is firmly grounded with mathematical motivation and proofs.
|
null | false
| null |
Is a life without meaning, a life of pessimism?
|
The idea of meaninglessness and absurdity of life was widely explored by many scholars, one being Albert Camus.
In accordance with his philosophical theory of absurdism the universe’s irrationality and randomness indicates that it does not posses a specific purpose. Therefore, one should abandon the search for meaning and embrace the absurdity of life. According to Camus, the lack of meaning in our existence is not negative, simply neutral. Without innate meaning, one is free to do with their life as they wish so the notion of a meaningless life isn’t objectively a useless life, nor is it a negative one.
|
|
null | false
| 388
|
The current coverage of the political landscape in the press and in social media has led to an unprecedented situation. Like never before, a statement in an interview, a press release, a blog note, or a tweet can spread almost instantaneously and reach the public in no time. This proliferation speed has left little time for double-checking claims against the facts, which has proven critical in politics, e.g., during the 2016 presidential campaign in the USA, which was arguably impacted by fake news in social media and by false claims.
Investigative journalists and volunteers have been working hard trying to get to the root of a claim and to present solid evidence in favor or against it. Manual fact-checking has proven very time-consuming, and thus automatic methods have been proposed as a way to speed-up the process. For instance, there has been work on checking the factuality/credibility of a claim, of a news article, or of an information source BIBREF0, BIBREF1, BIBREF2, BIBREF3, BIBREF4, BIBREF5, BIBREF6, BIBREF7. However, less attention has been paid to other steps of the fact-checking pipeline, which is shown in Figure FIGREF1.
The process starts when a document is made public. First, an intrinsic analysis is carried out in which check-worthy text fragments are identified. Then, other documents that might support or rebut a claim in the document are retrieved from various sources. Finally, by comparing a claim against the retrieved evidence, a system can determine whether the claim is likely true or likely false. For instance, BIBREF8 do this on the basis of a knowledge graph derived from Wikipedia. The outcome could then be presented to a human expert for final judgment.
In this paper, we focus on the first step: predicting check-worthiness of claims. Our contributions can be summarized as follows:
New dataset: We build a new dataset of manually-annotated claims, extracted from the 2016 US presidential and vice-presidential debates, which we gathered from nine reputable sources such as CNN, NPR, and PolitiFact, and which we release to the research community.
Modeling the context: We develop a novel approach for automatically predicting which claims should be prioritized for fact-checking, based on a rich input representation. In particular, we model not only the textual content, but also the context: how the target claim relates to the current segment, to neighboring segments and sentences, and to the debate as a whole, and also how the opponents and the public react to it.
State-of-the-art results: We achieve state-of-the-art results, outperforming a strong rivaling system by a margin, while also demonstrating that this improvement is due primarily to our modeling of the context.
We model the problem as a ranking task, and we train both Support Vector Machines (SVM) and Feed-forward Neural Networks (FNN) obtaining state-of-the-art results. We also analyze the relevance of the specific feature groups and we show that modeling the context yields a significant boost in performance. Finally, we also analyze whether we can learn to predict which facts are check-worthy with respect to each of the individual media sources, thus capturing their biases. It is worth noting that while trained on political debates, many features of our model can be potentially applied to other kinds of information sources, e.g., interviews and news.
The rest of the paper is organized as follows: Section SECREF2 discusses related work. Section SECREF3 describes the process of gathering and annotating our political debates dataset. Section SECREF4 presents our supervised approach to predicting fact-checking worthiness, including the explanation of the model and the information sources we use. Section SECREF5 presents the evaluation setup and discusses the results. Section SECREF6 provides further analysis. Finally, Section SECREF7 presents the conclusions and outlines some possible directions for future research.
State-of-the-art results: We achieve state-of-the-art results, outperforming a strong rivaling system by a margin, while also demonstrating that this improvement is due primarily to our modeling of the context.
|
Does the authors' system outperform a strong rivaling system by a margin?
|
Yes, it does.
|
null | false
| null |
What is the best food in the world?
|
Humans love their food, don’t they? Why is that? Well, humans are born in places where there is a local cuisine that they grow up with and love. Eating this food established an association with a comforting experience and this is why most nations claim that the best food is in their country. However, there seems to be a global understanding that Italian, French, Thai and Japanese cuisines are some of the best in the world. The reason for this, apart from top quality genuine ingredients and local cultural importance, can be the positive association attached to the cuisine by mass media or personal experiences whilst on holiday.
|
|
null | false
| null |
Shauraseni Prakrit (Sanskrit: शौरसेनी प्राकृत, Śaurasenī Prākṛta) was a Middle Indo-Aryan language and a Dramatic Prakrit. Shauraseni was the chief language used in drama in medieval northern India. Most of the material in this language originates from the 3rd to 10th centuries, though it was probably a spoken vernacular around the 2nd century BCE in the ancient state of Surasena. Among the Prakrits, Shauraseni is said to be the one most closely related to Classical Sanskrit in that it "is derived from the Old Indian Indo-Aryan dialect of the Madhyadeśa on which Classical Sanskrit was mainly based.": 3–4 Its descendants include the languages of the Hindi Belt, the Central Zone of modern Indo-Aryan or Hindi languages, the standard registers of the Hindustani language based on the Delhi dialect.
|
What kind of language is Shauraseni Prakrit
|
Shauraseni Prakrit was a Middle Indo-Aryan language and a Dramatic Prakrit. Shauraseni was the chief language used in drama in medieval northern India. Most of the material in this language originates from the 3rd to 10th centuries, though it was probably a spoken vernacular around the 2nd century BCE in the ancient state of Surasena.
Among the Prakrits, Shauraseni is said to be the one most closely related to Classical Sanskrit in that it is derived from the Old Indian Indo-Aryan dialect of the Madhyadesa on which Classical Sanskrit was mainly based. Its descendants include the languages of the Hindi Belt, the Central Zone of modern Indo-Aryan or Hindi languages, the standard registers of the Hindustani language based on the Delhi dialect.
|
null | false
| null |
How do you treat a dog diarrhea
|
You can treat a dog diarrhea by letting the dog skip one meal and give ground turkey and rice plus one packet of probiotics to the dog. You can add some veggie if you see improvement.
|
|
null | false
| null |
The speaker of the House of Commons (French: président de la Chambre des communes) is the presiding officer of the lower house of the Parliament of Canada. A member of Parliament (MP), they are elected at the beginning of each new parliament by fellow MPs. The speaker's role in presiding over Canada's House of Commons is similar to that of speakers elsewhere in other countries that use the Westminster system.
The 37th and current speaker of the House of Commons is Anthony Rota, since December 5, 2019. The speaker with the longest tenure is Peter Milliken who was elected for four consecutive terms lasting 10 years, 124 days.
|
From the passage provided, extract the name of the 37th speaker of the House of Commons of Canada
|
The 37th and current speaker of the House of Commons is Anthony Rota
|
null | false
| null |
Portfolio optimization often takes place in two stages: optimizing weights of asset classes to hold, and optimizing weights of assets within the same asset class. An example of the former would be choosing the proportions placed in equities versus bonds, while an example of the latter would be choosing the proportions of the stock sub-portfolio placed in stocks X, Y, and Z
|
Extract the stages for portfolio optimization from https://en.wikipedia.org/wiki/Portfolio_optimization
|
- Stage 1 is optimizing weights of asset classes to hold, choose the allocation of equities and bonds
- Stage 2 is optimizing weights of assets within the same asset class, choose the proportions of each stock stickers in the stock portfolio
|
null | false
| null |
Identify which instrument is string or percussion: Lummi stick, Timple
|
Timple is string, Lummi stick is percussion.
|
|
null | false
| null |
What is Human Development Report (HDI) based on?
|
HDI is based on Life Expectancy, Literacy & Per Capita Income
|
|
null | false
| null |
Artificial neural networks (ANNs), usually simply called neural networks (NNs) or neural nets, are computing systems inspired by the biological neural networks that constitute animal brains. An ANN is based on a collection of connected units or nodes called artificial neurons, which loosely model the neurons in a biological brain. Each connection, like the synapses in a biological brain, can transmit a signal to other neurons. An artificial neuron receives signals then processes them and can signal neurons connected to it. The "signal" at a connection is a real number, and the output of each neuron is computed by some non-linear function of the sum of its inputs. The connections are called edges. Neurons and edges typically have a weight that adjusts as learning proceeds. The weight increases or decreases the strength of the signal at a connection. Neurons may have a threshold such that a signal is sent only if the aggregate signal crosses that threshold.
utomatically generate identifying characteristics from the examples that they process.Typically, neurons are aggregated into layers. Different layers may perform different transformations on their inputs. Signals travel from the first layer (the input layer), to the last layer (the output layer), possibly after traversing the layers multiple times.
|
What are artificial neural networks?
|
Neural networks are computer systems inspired by the biological neural networks that constitute animal brains
|
null | false
| null |
What items do you need to buy in preparation for the arrival of a baby
|
The arrival of a baby is an exciting time! Here is a list of some items you may want to get before its arrival:
- Crib or bassinet and bedding
- Baby monitor
- Car Seat
- Stroller
- Onesies and appropriate clothing
- Feeding bottles
- First aid kit
- Diapers and related creams and wipes
|
|
null | false
| null |
Classify each item as jewelry or clothing: ring, shirt, shorts, socks, necklace, bracelet
|
Ring: Jewelry
Shirt: Clothing
Shorts: Clothing
Socks: Clothing
Necklace: Jewelry
Bracelet: Jewelry
|
|
null | false
| null |
How do you install a wooden fence?
|
First, you would start by plotting out the line of the fence around your yard. You would want to make sure that you are not imposing on your neighbors yard, so it is always smart to get a survey done by the local government as well. When you are ready to start installing your fence, you would Dig holes using a post digger at intervals between 3 to 5 feet wide. He would then pour concrete into the holes and put the fence posts in the holes. After the posts are installed, and the concrete is dried, you would nail the wooden planks between the posts to make your fence. When you’re all done, it’s always nice to enjoy a cold beer if you are above the legal limit.
|
|
null | false
| null |
Please write a socia media post about why Leo Messi is the best soccer player in the world.
|
Messi is the best soccer player in the world because he has both the best skills and strongest mentality. His physical strength might not be the greatest in the world but he is able to leverage his body to perform amazing motions and skills on the field. His control of the ball is unmatchable and always give his opponents a hard time. In addition to that, Messi has a winner mentality that keeps him calm and focused especially in big matches. Will all those virtues, Messi has won countless trophies and awards so he is considered to be the best soccer player in the world.
|
|
null | false
| null |
Polly, a 51-year-old female, visits your fitness facility. She is dissatisfied with her body and is responding to a series of ads your facility uses to attract people interested in weight loss. What new behaviors and habits could she adopt to optimize weight control?
|
She can focus on three new behaviors: attitude rewiring, prioritizing consistency, and emotional control. First, she can adopt a new attitude towards health and fitness; instead of thinking of working out as a necessarily hard and unenjoyable task, she can start perceiving her workouts in a positive light. Instead of seeking to accomplish unrealistic goals, she can start small and focus on consistency, finding tasks that are both easy and enjoyable for her (such as walking a dog). Finally, because people tend to sabotage their good diets or workout plans due to distressing events, she needs to adopt emotional discipline, recognizing stressful situations and ensuring she is not acting purely out of stress.
|
|
null | false
| null |
Ivan Trotski is a Belarusian race walker.
|
Where is Ivan Trotski from?
|
Belarus
|
null | false
| null |
Time (stylized in all caps) is an American news magazine based in New York City. For nearly a century, it was published weekly, but starting in March 2020 it transitioned to every other week. It was first published in New York City on March 3, 1923, and for many years it was run by its influential co-founder, Henry Luce. A European edition (Time Europe, formerly known as Time Atlantic) is published in London and also covers the Middle East, Africa, and, since 2003, Latin America. An Asian edition (Time Asia) is based in Hong Kong. The South Pacific edition, which covers Australia, New Zealand, and the Pacific Islands, is based in Sydney.
Since 2018, Time has been owned by Salesforce founder Marc Benioff, who acquired it from Meredith Corporation. Benioff currently publishes the magazine through the company Time USA, LLC.
|
What is Time magazine?
|
An American news magazine, based in New York City since March 3,1923, published every week since and every other week since March 2020. Initially it was run by it's co-founder, Henry Luce but since 2018, it has been owned by Salesforce founder Marc Benioff, who publishes it through the company Time USA, LLC.
|
null | false
| null |
What are the lesser known Jewish holidays?
|
Most people know the Jewish holidays of Rosh Hashana, Yom Kippur, and Chanukah. Some of the lesser known Jewish holidays include the following: Sukkot, Shemini Atzeres, Simchat Torah, Tu B'Shvat, Purim, Passover (Pesach), Lag B'Omer, and Shavuot.
|
|
null | false
| null |
According to the Global Competitiveness Report of 2019, Morocco Ranked 32nd in the world in terms of Roads, 16th in Sea, 45th in Air and 64th in Railways. This gives Morocco the best infrastructure rankings in the African continent.
Modern infrastructure development, such as ports, airports, and rail links, is a top government priority. To meet the growing domestic demand, the Moroccan government invested more than $15 billion from 2010 to 2015 in upgrading its basic infrastructure.
Morocco has one of the best road systems on the continent. Over the past 20 years, the government has built approximately 1770 kilometers of modern roads, connecting most major cities via toll expressways. The Moroccan Ministry of Equipment, Transport, Logistics, and Water aims to build an additional 3380 kilometers of expressway and 2100 kilometers of highway by 2030, at an expected cost of $9.6 billion. It focuses on linking the southern provinces, notably the cities of Laayoune and Dakhla to the rest of Morocco.
In 2014, Morocco began the construction of the first high-speed railway system in Africa linking the cities of Tangiers and Casablanca. It was inaugurated in 2018 by the King following over a decade of planning and construction by Moroccan national railway company ONCF. It is the first phase of what is planned to eventually be a 1,500 kilometeres (930 mi) high-speed rail network in Morocco. An extension of the line to Marrakesh is already being planned.
Morocco also has the largest port in Africa and the Mediterranean called Tanger-Med, which is ranked the 18th in the world with a handling capacity of over 9 million containers. It is situated in the Tangiers free economic zone and serves as a logistics hub for Africa and the world.
|
What's the Moroccan infrastructure ?
|
Morocco has the best infrastructure in Africa.Morocco is ranked 32 in the world in terms of Roads, 16th in Sea , 45th in Air and 64 in Railways. A High-speed railway system was inaugurated in 2018. It's linking the cities of Tangiers and Casablanca. An extension of the line is already planned.
Morocco has also the largest port in Africa called Tanger Med. It's ranked 18th in the world.
|
null | false
| null |
Give me a bullet list of 6 different types of Christianity
|
- Catholicism
- Mormonism
- Presbyterian
- Methodist
- Baptist
- Calvinist
|
|
null | false
| null |
Explain to me why people decide to go vegan.
|
Vegans are primarily concerned with animal welfare. Humans do not need to eat animal products to be healthy, so using animals for food is unnecessary. Under capitalism, trillions of animals are raised in horrible conditions each year to be slaughtered to satisfy human taste buds. However, even if an animal is raised in conditions better than a factory farm, killing them is purely for taste since humans do not need to eat them to survive. Vegans view this as immoral and choose to not use animal products.
Additionally, meat has a high carbon footprint compared to eating plants. Beef is a particularly bad offender. Vast swathes of the rainforest are cleared away to make room to raise cows for slaughter. Many vegans also care about the environment and avoiding animal products is a way to contribute less damage to the environment.
|
|
null | false
| null |
The first European settlers in the present-day Austin were a group of Spanish friars who arrived from East Texas in July 1730. They established three temporary missions, La Purísima Concepción, San Francisco de los Neches and San José de los Nazonis, on a site by the Colorado River, near Barton Springs. The friars found conditions undesirable and relocated to the San Antonio River within a year of their arrival. Following Mexico's Independence from Spain, Anglo-American settlers began to populate Texas and reached present-day Central Texas by the 1830s. The site where Austin is located was surveyed by Edward Burleson in 1838, calling it Waterloo. It was incorporated in January, 1839, only months before selection as the site of the new capital, ending its existence. Early Austin resident and chronicler Frank Brown says the first and only settler in 1838 was Jacob Harrell who may have been living there already. Living in a tent with his family, he later built a cabin and small stockade near the mouth of Shoal Creek. In its short lifespan of less than two years the population of Waterloo grew to only about twelve people made up of four families.
|
What was the first name of the city that became Austin, Texas?
|
The site that became Austin was previously incorporated as Waterloo in 1839.
|
null | false
| null |
Should you buy a new or a used car?
|
You would firstly need to consider your budget and your appetite to borrow money, before you make a decision to buy a new or a used car.
|
|
null | false
| null |
What are the main reasons for the price of a good increasing or decreasing in a capitalist society.
|
Prices of goods in a capitalist society are continuously set by the levels of supply and demand for goods. Reasons the price of a good may increase include the supply of the good decreasing, the demand for the good increasing, or both. Alternatively, a price of a good may decrease if the supply of the good increases or the demand for the good decreases.
|
|
null | false
| null |
Can indoor houseplants live outside?
|
Houseplants that are considered invasive in your region should never be planted outside. In areas of North America this includes bamboo, some ivies such as English Ivy, snake plants, pothos, monstera, etc. If you’d like to enjoy these potentially invasive plants, do so indoors or in potted outdoor containers. Never discard potentially invasive houseplant species outdoors.
|
|
null | false
| null |
What is the boarding process at Southwest Airlines
|
Southwest Airlines offers a unique boarding process where the passenger is not given a specific seat, they are instead boarded in the order that they checked-in for the flight. Passengers are given the opportunity to upgrade their ticket, either through a loyalty program (Southwest Rewards) or by purchasing an Early Bird Check-In for a fee.
|
|
null | false
| null |
What should I expect from modern Data and AI Platform?
|
Improved Business Outcomes
Single repository for traditional and alternative data sources results in quick and easy to locate data and visualize data.
User friendly interface for SQL, Python, R, Java, and Scala users for data cleaning, data enrichment and feature engineering.
Easy to collaborate in data & AI lifecycle with data preparation, model development, model validation, model implementation, model execution, and model performance monitoring in a single platform.
Optimized and Well Governed Processes
Simple governance and security with one place to define controls on data tables/files, data pipelines, models and dashboards.
Simplifies business and IT processes with a single platform to achieve all required use cases.
Time-to-Market and ROI
High-performance computing results in faster data preparation, analytics and AI.
Well integrated platform with automations significantly reduces model calibration to decisioning.
Higher return on investment with improved business efficiency, and lower operational cost.
Competitive Imperative
Outcompete competitors by providing advanced models that can help financial institutions increase revenue.
Provide digital experience to your customer.
|
|
null | false
| null |
Which country has the largest population?
|
China and India seem to be competing for the title of "the country with the largest population". Currently the title is still with China with India expected to overtake soon. Both of these countries have population over 1.4 billion which is roughly 18% of the world population.
|
|
null | false
| 145
|
Language modeling is a probabilistic description of language phenomenon. It provides essential context to distinguish words which sound similar and therefore has one of the most useful applications in Natural Language Processing (NLP) especially in downstreaming tasks like Automatic Speech Recognition (ASR). Recurrent Neural Networks (RNN) especially Long Short Term Memory (LSTM) networks BIBREF0 have been the typical solution to language modeling which do achieve strong results. In spite of these results, their fundamental sequential computation constraint has restricted their use in the modeling of long-term dependencies in sequential data. To address these issues Transformer architecture was introduced. Transformers relies completely on an attention mechanism to form global dependencies between input and output. It also offers more parallelization and has achieved SOTA results in language modeling outperforming LSTM models BIBREF1.
In recent years,we have seen a lot of development based on this standard transformer models particularly on unsupervised pre-training(BIBREF2, BIBREF3, BIBREF4, BIBREF5, BIBREF6, BIBREF7 which have set state-of-the art results on multiple NLP benchmarks. One such model architecture has been the Bidirectional Encoder Representations from Transformers (BERT) model which uses a deep bidirectional transformer architecture.
Another architecture of interest would be the Transformer-XL, which introduces the notion of recurrence in a self-attention model.
The primary research focus though has been mostly on English language for which abundant data is present. It is interesting to see the performance of these models for an agglutinative language like Finnish, which is morphologically richer than English.
In this project, we explore the implementation of Transformer-based models (BERT and Transformer-XL) in language modeling for Finnish. We will use the same training data as in BIBREF8 so that we can do fair comparisons with the performance of the LSTM models. Also, as the BERT model is a bi-directional transformer, we will have to approximate the conditional probabilities given a sequence of words. We also experiment with using sub-word units with Transformer-XL to cope with the large vocabulary problems associated with the Finnish Language. With smaller units, the modeled sequences are longer, and we hope that the recursive XL architecture can allow us to still model long term effects. To the best of our knowledge this is the first work with the Finnish language to use the following:
Approximation of perplexity using a BERT architecture
Using Transformer-XL architecture with sub-word units.
Comparison of Transformer and LSTM models as language models in the same comparable settings with an agglutinative language.
Recurrent Neural Networks (RNN) especially Long Short Term Memory (LSTM) networks [1] have been the typical solution to language modeling which do achieve strong results. In spite of these results, their fundamental sequential computation constraint has restricted their use in the modeling of long-term dependencies in sequential data. To address these issues Transformer architecture was introduced.
|
Why transformer architecture was introduced?
|
To address these issues of RNN especially LSTM networks that their fundamental sequential computation constraint has restricted their use in the modeling of long-term dependencies in sequential data.
|
1703.02507
| false
| null |
FLOAT SELECTED: Table 1: Comparison of the performance of different models on different supervised evaluation tasks. An underline indicates the best performance for the dataset. Top 3 performances in each data category are shown in bold. The average is calculated as the average of accuracy for each category (For MSRP, we take the accuracy). )
We use a standard set of supervised as well as unsupervised benchmark tasks from the literature to evaluate our trained models, following BIBREF16 . The breadth of tasks allows to fairly measure generalization to a wide area of different domains, testing the general-purpose quality (universality) of all competing sentence embeddings. For downstream supervised evaluations, sentence embeddings are combined with logistic regression to predict target labels. In the unsupervised evaluation for sentence similarity, correlation of the cosine similarity between two embeddings is compared to human annotators.
Downstream Supervised Evaluation. Sentence embeddings are evaluated for various supervised classification tasks as follows. We evaluate paraphrase identification (MSRP) BIBREF25 , classification of movie review sentiment (MR) BIBREF26 , product reviews (CR) BIBREF27 , subjectivity classification (SUBJ) BIBREF28 , opinion polarity (MPQA) BIBREF29 and question type classification (TREC) BIBREF30 . To classify, we use the code provided by BIBREF22 in the same manner as in BIBREF16 . For the MSRP dataset, containing pairs of sentences INLINEFORM0 with associated paraphrase label, we generate feature vectors by concatenating their Sent2Vec representations INLINEFORM1 with the component-wise product INLINEFORM2 . The predefined training split is used to tune the L2 penalty parameter using cross-validation and the accuracy and F1 scores are computed on the test set. For the remaining 5 datasets, Sent2Vec embeddings are inferred from input sentences and directly fed to a logistic regression classifier. Accuracy scores are obtained using 10-fold cross-validation for the MR, CR, SUBJ and MPQA datasets. For those datasets nested cross-validation is used to tune the L2 penalty. For the TREC dataset, as for the MRSP dataset, the L2 penalty is tuned on the predefined train split using 10-fold cross-validation, and the accuracy is computed on the test set.
We propose a new unsupervised model, Sent2Vec, for learning universal sentence embeddings. Conceptually, the model can be interpreted as a natural extension of the word-contexts from C-BOW BIBREF0 , BIBREF1 to a larger sentence context, with the sentence words being specifically optimized towards additive combination over the sentence, by means of the unsupervised objective function.
The ParagraphVector DBOW model BIBREF14 is a log-linear model which is trained to learn sentence as well as word embeddings and then use a softmax distribution to predict words contained in the sentence given the sentence vector representation. They also propose a different model ParagraphVector DM where they use n-grams of consecutive words along with the sentence vector representation to predict the next word.
BIBREF16 propose a Sequential (Denoising) Autoencoder, S(D)AE. This model first introduces noise in the input data: Firstly each word is deleted with probability INLINEFORM0 , then for each non-overlapping bigram, words are swapped with probability INLINEFORM1 . The model then uses an LSTM-based architecture to retrieve the original sentence from the corrupted version. The model can then be used to encode new sentences into vector representations. In the case of INLINEFORM2 , the model simply becomes a Sequential Autoencoder. BIBREF16 also propose a variant (S(D)AE + embs.) in which the words are represented by fixed pre-trained word vector embeddings.
The SkipThought model BIBREF22 combines sentence level models with recurrent neural networks. Given a sentence INLINEFORM0 from an ordered corpus, the model is trained to predict INLINEFORM1 and INLINEFORM2 .
FastSent BIBREF16 is a sentence-level log-linear bag-of-words model. Like SkipThought, it uses adjacent sentences as the prediction target and is trained in an unsupervised fashion. Using word sequences allows the model to improve over the earlier work of paragraph2vec BIBREF14 . BIBREF16 augment FastSent further by training it to predict the constituent words of the sentence as well. This model is named FastSent + AE in our comparisons.
In a very different line of work, C-PHRASE BIBREF20 relies on additional information from the syntactic parse tree of each sentence, which is incorporated into the C-BOW training objective.
Compared to our approach, Siamese C-BOW BIBREF23 shares the idea of learning to average word embeddings over a sentence. However, it relies on a Siamese neural network architecture to predict surrounding sentences, contrasting our simpler unsupervised objective.
FLOAT SELECTED: Table 2: Unsupervised Evaluation Tasks: Comparison of the performance of different models on Spearman/Pearson correlation measures. An underline indicates the best performance for the dataset. Top 3 performances in each data category are shown in bold. The average is calculated as the average of entries for each correlation measure.
In Tables TABREF18 and TABREF19 , we compare our results with those obtained by BIBREF16 on different models. Table TABREF21 in the last column shows the dramatic improvement in training time of our models (and other C-BOW-inspired models) in contrast to neural network based models. All our Sent2Vec models are trained on a machine with 2x Intel Xeon E5 INLINEFORM0 2680v3, 12 cores @2.5GHz.
Along with the models discussed in Section SECREF3 , this also includes the sentence embedding baselines obtained by simple averaging of word embeddings over the sentence, in both the C-BOW and skip-gram variants. TF-IDF BOW is a representation consisting of the counts of the 200,000 most common feature-words, weighed by their TF-IDF frequencies. To ensure coherence, we only include unsupervised models in the main paper. Performance of supervised and semi-supervised models on these evaluations can be observed in Tables TABREF29 and TABREF30 in the supplementary material.
FLOAT SELECTED: Table 1: Comparison of the performance of different models on different supervised evaluation tasks. An underline indicates the best performance for the dataset. Top 3 performances in each data category are shown in bold. The average is calculated as the average of accuracy for each category (For MSRP, we take the accuracy). )
We use a standard set of supervised as well as unsupervised benchmark tasks from the literature to evaluate our trained models, following BIBREF16
Sentence embeddings are evaluated for various supervised classification tasks as follows. We evaluate paraphrase identification (MSRP) BIBREF25 , classification of movie review sentiment (MR) BIBREF26 , product reviews (CR) BIBREF27 , subjectivity classification (SUBJ) BIBREF28 , opinion polarity (MPQA) BIBREF29 and question type classification (TREC) BIBREF30 .
We propose a new unsupervised model, Sent2Vec, for learning universal sentence embeddings.
The ParagraphVector DBOW model BIBREF14 is a log-linear model which is trained to learn sentence as well as word embeddings and then use a softmax distribution to predict words contained in the sentence given the sentence vector representation.
They also propose a different model ParagraphVector DM where they use n-grams of consecutive words along with the sentence vector representation to predict the next word.
BIBREF16 propose a Sequential (Denoising) Autoencoder, S(D)AE.
The SkipThought model BIBREF22 combines sentence level models with recurrent neural networks.
FastSent BIBREF16 is a sentence-level log-linear bag-of-words model.
In a very different line of work, C-PHRASE BIBREF20 relies on additional information from the syntactic parse tree of each sentence, which is incorporated into the C-BOW training objective.
Compared to our approach, Siamese C-BOW BIBREF23 shares the idea of learning to average word embeddings over a sentence.
FLOAT SELECTED: Table 2: Unsupervised Evaluation Tasks: Comparison of the performance of different models on Spearman/Pearson correlation measures. An underline indicates the best performance for the dataset. Top 3 performances in each data category are shown in bold. The average is calculated as the average of entries for each correlation measure.
In Tables TABREF18 and TABREF19 , we compare our results with those obtained by BIBREF16 on different models.
Along with the models discussed in Section SECREF3 , this also includes the sentence embedding baselines obtained by simple averaging of word embeddings over the sentence, in both the C-BOW and skip-gram variants. TF-IDF BOW is a representation consisting of the counts of the 200,000 most common feature-words, weighed by their TF-IDF frequencies
In a very different line of work, C-PHRASE BIBREF20 relies on additional information from the syntactic parse tree of each sentence, which is incorporated into the C-BOW training objective.
|
Which other unsupervised models are used for comparison?
|
The answers are shown as follows:
* Sequential (Denoising) Autoencoder
* TF-IDF BOW
* SkipThought
* FastSent
* Siamese C-BOW
* C-BOW
* C-PHRASE
* ParagraphVector
|
null | false
| null |
Is there a World Taekwondo Grand Prix
|
The World Taekwondo Grand Prix is a taekwondo competition introduced by the World Taekwondo Federation in 2013 to provide a homogeneous system for qualification to the Olympic taekwondo tournament. It consists of four competitions per year in each Olympic weight category event. Olympic events occur at approximately half the weight classes as WTF-organised tournaments.
|
|
null | false
| 244
|
In Yugoslavia, organized effort in machine translation started in 1959, but the first individual effort was made by Vladimir Matković from the Institute for Telecommunications in Zagreb in 1957 in his PhD thesis on entropy in the Croatian language BIBREF10. The main research group in machine translation was formed in 1958, at the Circle for Young Linguists in Zagreb, initiated by a young linguist Bulcsu Laszlo, who graduated in Russian language, Southern Slavic languages and English language and literature at the University of Zagreb in 1952. The majority of the group members came from different departments of the Faculty of Humanities and Social Sciences of the University of Zagreb, with several individuals from other institutions. The members from the Faculty of Humanities and Social Sciences were: Svetozar Petrović (Department of Comparative Literature), Stjepan Babić (Department of Serbo-Croatian Language and Literature), Krunoslav Pranjić (Department of Serbo-Croatian Language and Literature), Željko Bujas (Department of English Language and Literature), Malik Mulić (Department of Russian Language and Literature) and Bulcsu Laszlo (Department of Comparative Slavistics). The members of the research group from outside the Faculty of Humanities and Social Sciences were: Božidar Finka (Institute for Language of the Yugoslav Academy of Sciences and Arts), Vladimir Vranić (Center for Numerical Research of the Yugoslav Academy of Sciences and Arts), Vladimir Matković (Institute for Telecommunications), Vladimir Muljević (Institute for Regulatory and Signal Devices) BIBREF10.
Laszlo and Petrović BIBREF11 also commented on the state of the art of the time, noting the USA prototype efforts from 1954 and the publication of a collection of research papers in 1955 as well as the USSR efforts starting from 1955 and the UK prototype from 1956. They do not detail or cite the articles they mention. However, the fact that they referred to them in a text published in 1959 (probably prepared for publishing in 1958, based on BIBREF11, where Laszlo and Petrović described that the group had started its work in 1958) leads us to the conclusion that the poorly funded Croatian research was lagging only a couple of years behind the research of the superpowers (which invested heavily in this effort). Another interesting moment, which they delineated in BIBREF11, is that the group soon discovered that some experimental work had already been done in 1957 at the Institute of Telecommunications (today a part of the Faculty of Electrical Engineering and Computing at the University of Zagreb) by Vladimir Matković. Because of this, they decided to include him in the research group of the Faculty of Humanities and Social Sciences at the University of Zagreb. The work done by Matković was documented in his doctoral dissertation but remained unpublished until 1959.
The Russian machine translation pioneer Andreev expressed hope that the Yugoslav (Croatian) research group could create a prototype, but sadly, due to the lack of federal funding, this never happened BIBREF10. Unlike their colleagues in the USA and the USSR, Laszlo’s group had to manage without an actual computer (which is painfully obvious in BIBREF12), and the results remained mainly theoretical. Appealing probably to the political circles of the time, Laszlo and Petrović note that, although it sounds strange, research in computational linguistics is mainly a top-priority military effort in other countries BIBREF11. There is a quote from BIBREF10 which perhaps best delineates the optimism and energy that the researchers in Zagreb had:
"[...] The process of translation has to mechanicalized as soon as possible, and this is only possible if a competent, fast and inexhaustible machine which could inherit the translation task is created, even if just schematic. The machine needs to think for us. If machines help humans in physical tasks, why would they not help them in mental tasks with their mechanical memory and automated logic" (p. 118).
In Yugoslavia, organized effort in machine translation started in 1959, but the first individual effort was made by Vladimir Matkovi´c from the Institute for Telecommunications in Zagreb in 1957 in his PhD thesis on entropy in the Croatian language [7].
|
Who made the first individual effort in Yugoslavia?
|
The first individual effort was made by Vladimir Matkovi´c from the Institute for Telecommunications in Zagreb in 1957 in his PhD thesis on entropy in the Croatian language.
|
null | false
| null |
The classis horror the Shining exterior shots were shot at this location:
|
the Spirit Lake Lodge in Colorado, the Sunriver Resort in Oregon, the TImberline Lodge in Oregon, the Overlook Lodge in Oregon, or the Spirit Lake Lodge in Colorado.
|
|
1909.13375
| false
| null |
Table TABREF25 shows the results on DROP's test set, with our model being the best overall as of the time of writing, and not just on multi-span questions.
FLOAT SELECTED: Table 3. Comparing test and development set results of models from the official DROP leaderboard
Table TABREF25 shows the results on DROP's test set, with our model being the best overall as of the time of writing, and not just on multi-span questions.
FLOAT SELECTED: Table 3. Comparing test and development set results of models from the official DROP leaderboard
|
What is the performance of proposed model on entire DROP dataset?
|
The proposed model achieves EM 77,63 and F1 80,73 on the test and EM 76,95 and F1 80,25 on the dev
|
null | false
| null |
Who won highest number of Men's singles Wimbledon titles?
|
Roger Federer won highest number of Wimbledon titles (8 Titles)
|
|
1909.00578
| false
| null |
Methods ::: Baselines ::: BiGRU s with attention:
This is very similar to Sum-QE but now $\mathcal {E}$ is a stack of BiGRU s with self-attention BIBREF21, instead of a BERT instance. The final summary representation ($h$) is the sum of the resulting context-aware token embeddings ($h = \sum _i a_i h_i$) weighted by their self-attention scores ($a_i$). We again have three flavors: one single-task (BiGRU-ATT-S-1) and two multi-task (BiGRU-ATT-M-1 and BiGRU-ATT-M-5).
Methods ::: Baselines ::: ROUGE:
This baseline is the ROUGE version that performs best on each dataset, among the versions considered by BIBREF13. Although ROUGE focuses on surface similarities between peer and reference summaries, we would expect properties like grammaticality, referential clarity and coherence to be captured to some extent by ROUGE versions based on long $n$-grams or longest common subsequences.
Methods ::: Baselines ::: Language model (LM):
For a peer summary, a reasonable estimate of $\mathcal {Q}1$ (Grammaticality) is the perplexity returned by a pre-trained language model. We experiment with the pre-trained GPT-2 model BIBREF22, and with the probability estimates that BERT can produce for each token when the token is treated as masked (BERT-FR-LM). Given that the grammaticality of a summary can be corrupted by just a few bad tokens, we compute the perplexity by considering only the $k$ worst (lowest LM probability) tokens of the peer summary, where $k$ is a tuned hyper-parameter.
Methods ::: Baselines ::: Next sentence prediction:
BERT training relies on two tasks: predicting masked tokens and next sentence prediction. The latter seems to be aligned with the definitions of $\mathcal {Q}3$ (Referential Clarity), $\mathcal {Q}4$ (Focus) and $\mathcal {Q}5$ (Structure & Coherence). Intuitively, when a sentence follows another with high probability, it should involve clear referential expressions and preserve the focus and local coherence of the text. We, therefore, use a pre-trained BERT model (BERT-FR-NS) to calculate the sentence-level perplexity of each summary:
where $p(s_i|s_{i-1})$ is the probability that BERT assigns to the sequence of sentences $\left< s_{i-1}, s \right>$, and $n$ is the number of sentences in the peer summary.
Methods ::: Baselines ::: BiGRU s with attention:
This is very similar to Sum-QE but now $\mathcal {E}$ is a stack of BiGRU s with self-attention BIBREF21, instead of a BERT instance. The final summary representation ($h$) is the sum of the resulting context-aware token embeddings ($h = \sum _i a_i h_i$) weighted by their self-attention scores ($a_i$). We again have three flavors: one single-task (BiGRU-ATT-S-1) and two multi-task (BiGRU-ATT-M-1 and BiGRU-ATT-M-5).
Methods ::: Baselines ::: ROUGE:
This baseline is the ROUGE version that performs best on each dataset, among the versions considered by BIBREF13. Although ROUGE focuses on surface similarities between peer and reference summaries, we would expect properties like grammaticality, referential clarity and coherence to be captured to some extent by ROUGE versions based on long $n$-grams or longest common subsequences.
Methods ::: Baselines ::: Language model (LM):
For a peer summary, a reasonable estimate of $\mathcal {Q}1$ (Grammaticality) is the perplexity returned by a pre-trained language model. We experiment with the pre-trained GPT-2 model BIBREF22, and with the probability estimates that BERT can produce for each token when the token is treated as masked (BERT-FR-LM). Given that the grammaticality of a summary can be corrupted by just a few bad tokens, we compute the perplexity by considering only the $k$ worst (lowest LM probability) tokens of the peer summary, where $k$ is a tuned hyper-parameter.
Methods ::: Baselines ::: Next sentence prediction:
BERT training relies on two tasks: predicting masked tokens and next sentence prediction. The latter seems to be aligned with the definitions of $\mathcal {Q}3$ (Referential Clarity), $\mathcal {Q}4$ (Focus) and $\mathcal {Q}5$ (Structure & Coherence). Intuitively, when a sentence follows another with high probability, it should involve clear referential expressions and preserve the focus and local coherence of the text. We, therefore, use a pre-trained BERT model (BERT-FR-NS) to calculate the sentence-level perplexity of each summary:
where $p(s_i|s_{i-1})$ is the probability that BERT assigns to the sequence of sentences $\left< s_{i-1}, s \right>$, and $n$ is the number of sentences in the peer summary.
|
What simpler models do they look at?
|
BiGRUs with attention, ROUGE, Language model, and next sentence prediction
|
null | false
| null |
What is the loudest animal on Earth?
|
The sperm whale.
|
|
2001.05970
| false
| null |
In this study, we limit the sample size to the followers identified as English speakers in the U.S. News Top 200 National Universities. We utilize the Jefferson-Henrique script, a web scraper designed for Twitter to retrieve a total of over 300,000 #MeToo tweets from October 15th, when Alyssa Milano posted the inceptive #MeToo tweet, to November 15th of 2017 to cover a period of a month when the trend was on the rise and attracting mass concerns. Since the lists of the followers of the studied colleges might overlap and many Twitter users tend to reiterate other's tweets, simply putting all the data collected together could create a major redundancy problem. We extract unique users and tweets from the combined result set to generate a dataset of about 60,000 unique tweets, pertaining to 51,104 unique users.
We extract unique users and tweets from the combined result set to generate a dataset of about 60,000 unique tweets, pertaining to 51,104 unique users.
|
How many tweets are explored in this paper?
|
The answers are shown as follows:
* 60,000
|
null | false
| null |
Credit Suisse Group AG (French pronunciation: [kʁe.di sɥis], lit. 'Swiss Credit') is a global investment bank and financial services firm founded and based in Switzerland. Headquartered in Zürich, it maintains offices in all major financial centers around the world and provides services in investment banking, private banking, asset management, and shared services. It is known for strict bank–client confidentiality and banking secrecy. The Financial Stability Board considers it to be a global systemically important bank. Credit Suisse is also a primary dealer and Forex counterparty of the Federal Reserve in the United States.
Credit Suisse was founded in 1856 to fund the development of Switzerland's rail system. It issued loans that helped create Switzerland's electrical grid and the European rail system. In the 1900s, it began shifting to retail banking in response to the elevation of the middle class and competition from fellow Swiss banks UBS and Julius Bär. Credit Suisse partnered with First Boston in 1978 before buying a controlling share of the bank in 1988. From 1990 to 2000, the company purchased institutions such as Winterthur Group, Swiss Volksbank, Swiss American Securities Inc. (SASI), and Bank Leu. The biggest institutional shareholders of Credit Suisse include the Saudi National Bank (9.88%), the Qatar Investment Authority and BlackRock (about 5% each), Dodge & Cox, Norges Bank and the Saudi Olayan Group.
The company was one of the least affected banks during the global financial crisis, but afterwards began shrinking its investment business, executing layoffs and cutting costs. The bank was at the center of multiple international investigations for tax avoidance which culminated in a guilty plea and the forfeiture of US$2.6 billion in fines from 2008 to 2012. By the end of 2022, Credit Suisse had approximately CHF 1.3 trillion in assets under management.
|
From the passage note down the biggest institutional shareholders of Credit Suisse. List the results in comma separated format.
|
Saudi National Bank, Qatar Investment Authority, BlackRock, Dodge & Cox, Norges Bank, Saudi Olayan Group
|
null | false
| null |
Around the beginning of the 20th century, the DTES was Vancouver's political, cultural and retail centre. Over several decades, the city centre gradually shifted westwards, and the DTES became a poor neighbourhood, although relatively stable. In the 1980s, the area began a rapid decline due to several factors, including an influx of hard drugs, policies that pushed sex work and drug-related activity out of nearby areas, and the cessation of federal funding for social housing. By 1997, an epidemic of HIV infection and drug overdoses in the DTES led to the declaration of a public health emergency. As of 2018, critical issues include opioid overdoses, especially those involving the drug fentanyl; decrepit and squalid housing; a shortage of low-cost rental housing; and mental illness, which often co-occurs with addiction.
|
Based on the text, please give me a brief summary of the history of the DTES neighbourhood in Vancouver
|
The DTES has transformed from the focal point of Vancouver’s cultural and commercial life at the start of the 20th Century to an impoverished neighbourhood with many socio-economic problems by the 1980s. Since the 1990s, the DTES has been home to a public health emergency due to drug overdoses, HIV infection, a housing crisis, and mental health problems.
|
null | false
| null |
Is Gillian Anderson an English actress?
|
No. Gillian Anderson is American. She was born in Chicago, Illinois. However, she spent many years living in England as a child and was able to develop an English accent. She is considered bidialectal and can switch between American and British accents. She currently lives in London.
|
|
null | false
| null |
How old do you have to be to get your pilot's license in the United States.
|
The answer varies based on type of license and type of aircraft. But the most common Private Pilot License for fixed wing aircraft has a minimum age of 17 years.
|
|
null | false
| null |
Classify each of the following universities as being in the ACC or SEC: Ole Miss, LSU, FSU, Clemson, UGA, Miami
|
Ole Miss: SEC
LSU: SEC
FSU: ACC
Clemson: ACC
UGA: SEC
Miami: ACC
|
|
null | false
| null |
Why do hindus worship idols?
|
Religion is man made. History teaches us that Religion was invented to separate humans and create a fear. Religion and Faith are different. Hinduism is not a religion but on the contrary it is a way of life. There is no one leader or master in hinduism like in christianity or islam and it is totally inclusive. Anyone can call themselves a hindu and no one will balk at them. The qualities we admire in humans that we wish to pursue cannot be easily manifested. Thus an idol sometimes takes it's place. As an example the monkey god is known for courage. So anytime you need to stand up you pray to the monkey god. The elephant god is for good beginnings. So start your day with a prayer to elephant god. Thus you have 1000's of idols each representing a specific character in humans.
|
|
null | false
| null |
Food historian Lois Ellen Frank calls potatoes, tomatoes, corn, beans, squash, chili, cacao, and vanilla the "magic eight" ingredients that were found and used only in the Americas before 1492 and were taken via the Columbian Exchange back to the Old World, dramatically transforming the cuisine there. According to Frank,
If we deconstruct that these foods were inherently native, then that means that the Italians didn't have the tomato, the Irish didn't have the potato, half the British National Dish—Fish and Chips—didn't exist. The Russians didn't have the potato, nor did they have vodka from the potato. There were no chiles in any Asian cuisine anywhere in the world, nor were there any chiles in any East Indian cuisine dishes, including curries. And the French had no confection using either vanilla or chocolate. So the Old World was a completely different place.
|
Given this paragraph, what key crops were not available in Europe before the colonization of the New World?
|
Potatoes, tomatoes, corn, beans, squash, chili, cacao, and vanilla were not available in the Old World prior to the European discovery of the Americas
|
null | false
| 383
|
Transfer learning allows to use a neural network trained for one task to be used in another domain. We use transfer learning to classify patients and healthy speakers in three different languages. The CNN architecture described before is used to train a CNN with utterances from one language. Then, the pre-trained model is used as a base to initialize two different models with the remaining languages. Figure FIGREF11 summarizes this procedure.
Transfer learning allows to use a neural network trained for one task to be used in another domain. We use transfer learning to classify patients and healthy speakers in three different languages.
|
What do they use transfer learning to do?
|
To classify patients and healthy speakers in three different languages.
|
null | false
| null |
Who was the US president during the 9/11 attacks?
|
President George W. Bush
|
|
1811.01734
| false
| null |
Data set. For the cross-domain polarity classification experiments, we use the second version of Multi-Domain Sentiment Dataset BIBREF0 . The data set contains Amazon product reviews of four different domains: Books (B), DVDs (D), Electronics (E) and Kitchen appliances (K). Reviews contain star ratings (from 1 to 5) which are converted into binary labels as follows: reviews rated with more than 3 stars are labeled as positive, and those with less than 3 stars as negative. In each domain, there are 1000 positive and 1000 negative reviews.
For the cross-domain polarity classification experiments, we use the second version of Multi-Domain Sentiment Dataset BIBREF0 . The data set contains Amazon product reviews of four different domains: Books (B), DVDs (D), Electronics (E) and Kitchen appliances (K).
|
What domains are contained in the polarity classification dataset?
|
The answers are shown as follows:
* Books
* DVDs
* Electronics
* Kitchen appliances
|
1911.13066
| false
| null |
FLOAT SELECTED: Table 1. Description of class label along with distribution of each class (in %) in the acquired dataset
FLOAT SELECTED: Table 1. Description of class label along with distribution of each class (in %) in the acquired dataset
|
What is the 12 class bilingual text?
|
Appreciation, Satisfied, Peripheral complaint, Demanded inquiry, Corruption, Lagged response, Unresponsive, Medicine payment, Adverse behavior, Grievance ascribed and Obnoxious/irrelevant
|
null | false
| null |
Classify each of the following as a primary color or a secondary color
|
blue, yellow, green, red, purple, orange
|
|
null | false
| null |
Who is the prime minister of India
|
The prime minister of India is Narendra Modi who has been in power since 2014. He belongs to the Bharatiya Janata Party (BJP). He is the 14th prime minister of India.
|
|
null | false
| null |
Where should I go on vacation?
|
Well, that depends on your travel budget and how much time you want to spend, and the kind of vacation you like. Here are some recommendations for a vacation by the Time magazine: Dominica, Barcelona (Spain), Torres del Paine National Park (Chile), Ladakh (India), Churchill (Manitoba), Aarhus (Denmark), Roatán (Honduras), Aqaba (Jordan), Kyoto (Japan). These are just some recommendations of beautiful places to visit.
|
|
1810.10254
| false
| null |
FLOAT SELECTED: Table 3. Language Modeling Results (in perplexity).
UTF8gbsn The pointer-generator significantly outperforms the Seq2Seq with attention model by 3.58 BLEU points on the test set as shown in Table TABREF8 . Our language modeling result is given in Table TABREF9 . Based on the empirical result, adding generated samples consistently improve the performance of all models with a moderate margin around 10% in perplexity. After all, our proposed method still slightly outperforms the heuristic from linguistic constraint. In addition, we get a crucial gain on performance by adding syntax representation of the sequences.
FLOAT SELECTED: Table 3. Language Modeling Results (in perplexity).
Our language modeling result is given in Table TABREF9 .
|
What was their perplexity score?
|
Perplexity score 142.84 on dev and 138.91 on test
|
null | false
| 97
|
The LexVec BIBREF7 model factorizes the PPMI-weighted word-context co-occurrence matrix using stochastic gradient descent.
$$PPMI_{wc} = max(0, \log \frac{M_{wc} \; M_{**}}{ M_{w*} \; M_{*c} })$$ (Eq. 3)
where $M$ is the word-context co-occurrence matrix constructed by sliding a window of fixed size centered over every target word
$w$ in the subsampled BIBREF2 training corpus and incrementing cell $M_{wc}$ for every context word $c$ appearing within this window (forming a $(w,c)$ pair). LexVec adjusts the PPMI matrix using context distribution smoothing BIBREF3 .
With the PPMI matrix calculated, the sliding window process is repeated and the following loss functions are minimized for every observed $(w,c)$ pair and target word $w$ :
$$L_{wc} &= \frac{1}{2} (u_w^\top v_c - PPMI_{wc})^2 \\
L_{w} &= \frac{1}{2} \sum \limits _{i=1}^k{\mathbf {E}_{c_i \sim P_n(c)} (u_w^\top v_{c_i} - PPMI_{wc_i})^2 }$$ (Eq. 4)
where $u_w$ and $v_c$ are $d$ -dimensional word and context vectors. The second loss function describes how, for each target word, $k$ negative samples BIBREF2 are drawn from the smoothed context unigram distribution.
Given a set of subwords $S_w$ for a word $w$ , we follow fastText and replace $u_w$ in eq:lexvec2,eq:lexvec3 by $u^{\prime }_w$ such that:
$$u^{\prime }_w = \frac{1}{|S_w| + 1} (u_w + \sum _{s \in S_w} q_{hash(s)})$$ (Eq. 5)
such that a word is the sum of its word vector and its $d$ -dimensional subword vectors $q_x$ . The number of possible subwords is very large so the function $hash(s)$ hashes a subword to the interval $[1, buckets]$ . For OOV words,
$$u^{\prime }_w = \frac{1}{|S_w|} \sum _{s \in S_w} q_{hash(s)}$$ (Eq. 7)
We compare two types of subwords: simple n-grams (like fastText) and unsupervised morphemes. For example, given the word “cat”, we mark beginning and end with angled brackets and use all n-grams of length 3 to 6 as subwords, yielding $S_{\textnormal {cat}} = \lbrace \textnormal {$ $ ca, at$ $, cat} \rbrace $ . Morfessor BIBREF11 is used to probabilistically segment words into morphemes. The Morfessor model is trained using raw text so it is entirely unsupervised. For the word “subsequent”, we get $S_{\textnormal {subsequent}} = \lbrace \textnormal {$ $ sub, sequent$ $} \rbrace $ .
The Morfessor model is trained using raw text so it is entirely unsupervised.
|
Is Morfessor an entirely unsupervised model?
|
Yes.
|
null | false
| 170
|
Our first dataset contains articles from celebrity magazines People, UsWeekly, and E!News. We labeled each article for whether it was reporting on men, women, or neither/unknown. To do this, we first extracted the article's topic tags. Some of these tags referred to people, but others to non-people entities, such as “Gift Ideas” or “Health.” To distinguish between these types of tags, we queried each tag on Wikipedia and checked whether the top page result contained a “Born” entry in its infobox – if so, we concluded that the tag referred to a person.
Then, from the person's Wikipedia page, we determined their gender by checking whether the introductory paragraphs of the page contained more male or female pronouns. This method was simple but effective, since pronouns in the introduction almost always resolve to the subject of that page. In fact, on a sample of 80 tags that we manually annotated, we found that comparing pronoun counts predicted gender with perfect accuracy. Finally, if an article tagged at least one woman and did not tag any men, we labeled the article as Female; in the opposite case, we labeled it as Male.
Our second dataset contains reviews from RateMyProfessors (RMP), an online platform where students can review their professors. We included all 5,604 U.S. schools on RMP, and collected all reviews for CS professors at those schools. We labeled each review with the gender of the professor whom it was about, which we determined by comparing the count of male versus female pronouns over all reviews for that professor. This method was again effective, because the reviews are expressly written about a certain professor, so the pronouns typically resolve to that professor.
In addition to extracting the text of the articles or reviews, for each dataset we also collected various useful metadata. For the celebrity dataset, we recorded each article's timestamp and the name of the author, if available. Storing author names creates the potential to examine the relationship between the gender of the author and the gender of the subject, such as asking if there are differences between how women write about men and how men write about men. In this work, we did not yet pursue this direction because we wanted to begin with a simpler question of how gender is discussed: regardless of the gender of the authors, what is the content being put forth and consumed? Furthermore, we were unable to extract author gender in the professor dataset since the RMP reviews are anonymous. However, in future work, we may explore the influence of author gender in the celebrity dataset.
For the professor dataset, we captured metadata such as each review's rating, which indicates how the student feels about the professor on a scale of AWFUL to AWESOME. This additional variable in our data creates the option in future work to factor in sentiment; for example, we could study whether there are differences in language used when criticizing a female versus a male professor.
Our first dataset contains articles from celebrity magazines People, UsWeekly, and E!News.
|
What does the first dataset contain?
|
The first dataset contains articles from celebrity magazines People, UsWeekly, and E!News.
|
1809.10644
| false
| null |
In this paper, we use three data sets from the literature to train and evaluate our own classifier. Although all address the category of hateful speech, they used different strategies of labeling the collected data. Table TABREF5 shows the characteristics of the datasets.
Data collected by BIBREF3 , which we term the Sexist/Racist (SR) data set, was collected using an initial Twitter search followed by analysis and filtering by the authors and their team who identified 17 common phrases, hashtags, and users that were indicative of abusive speech. BIBREF4 collected the HATE dataset by searching for tweets using a lexicon provided by Hatebase.org. The final data set we used, which we call HAR, was collected by BIBREF9 ; we removed all retweets reducing the dataset to 20,000 tweets. Tweets were labeled as “Harrassing” or “Non-Harrassing”; hate speech was not explicitly labeled, but treated as an unlabeled subset of the broader “Harrassing” category BIBREF9 .
In this paper, we use three data sets from the literature to train and evaluate our own classifier.
Data collected by BIBREF3 , which we term the Sexist/Racist (SR) data set, was collected using an initial Twitter search followed by analysis and filtering by the authors and their team who identified 17 common phrases, hashtags, and users that were indicative of abusive speech. BIBREF4 collected the HATE dataset by searching for tweets using a lexicon provided by Hatebase.org. The final data set we used, which we call HAR, was collected by BIBREF9 ; we removed all retweets reducing the dataset to 20,000 tweets. Tweets were labeled as “Harrassing” or “Non-Harrassing”; hate speech was not explicitly labeled, but treated as an unlabeled subset of the broader “Harrassing” category BIBREF9 .
|
Which publicly available datasets are used?
|
The answers are shown as follows:
* BIBREF3
* BIBREF4
* BIBREF9
|
null | false
| 31
|
Single-document summarization is the task of generating a short summary for a given document. Ideally, the generated summaries should be fluent and coherent, and should faithfully maintain the most important information in the source document. purpleThis is a very challenging task, because it arguably requires an in-depth understanding of the source document, and current automatic solutions are still far from human performance BIBREF0 .
Single-document summarization can be either extractive or abstractive. Extractive methods typically pick sentences directly from the original document based on their importance, and form the summary as an aggregate of these sentences. Usually, summaries generated in this way have a better performance on fluency and grammar, but they may contain much redundancy and lack in coherence across sentences. In contrast, abstractive methods attempt to mimic what humans do by first extracting content from the source document and then produce new sentences that aggregate and organize the extracted information. Since the sentences are generated from scratch they tend to have a relatively worse performance on fluency and grammar. Furthermore, while abstractive summaries are typically less redundant, they may end up including misleading or even utterly false statements, because the methods to extract and aggregate information form the source document are still rather noisy.
In this work, we focus on extracting informative sentences from a given document (without dealing with redundancy), especially when the document is relatively long (e.g., scientific articles).
Most recent works on neural extractive summarization have been rather successful in generating summaries of short news documents (around 650 words/document) BIBREF1 by applying neural Seq2Seq models BIBREF2 . However when it comes to long documents, these models tend to struggle with longer sequences because at each decoding step, the decoder needs to learn to construct a context vector capturing relevant information from all the tokens in the source sequence BIBREF3 .
Long documents typically cover multiple topics. In general, the longer a document is, the more topics are discussed. As a matter of fact, when humans write long documents they organize them in chapters, sections etc.. Scientific papers are an example of longer documents and they follow a standard discourse structure describing the problem, methodology, experiments/results, and finally conclusions BIBREF4 .
To the best of our knowledge only one previous work in extractive summarization has explicitly leveraged section information to guide the generation of summaries BIBREF5 . However, the only information about sections fed into their sentence classifier is a categorical feature with values like Highlight, Abstract, Introduction, etc., depending on which section the sentence appears in.
In contrast, in order to exploit section information, in this paper we propose to capture a distributed representation of both the global (the whole document) and the local context (e.g., the section/topic) when deciding if a sentence should be included in the summary
Our main contributions are as follows: (i) In order to capture the local context, we are the first to apply LSTM-minus to text summarization. LSTM-minus is a method for learning embeddings of text spans, which has achieved good performance in dependency parsing BIBREF6 , in constituency parsing BIBREF7 , as well as in discourse parsing BIBREF8 . With respect to more traditional methods for capturing local context, which rely on hierarchical structures, LSTM-minus produces simpler models i.e. with less parameters, and therefore faster to train and less prone to overfitting. (ii) We test our method on the Pubmed and arXiv datasets and results appear to support our goal of effectively summarizing long documents. In particular, while overall we outperform the baseline and previous approaches only by a narrow margin on both datasets, the benefit of our method become much stronger as we apply it to longer documents. purpleFurthermore, in an ablation study to assess the relative contributions of the global and the local model we found that, rather surprisingly, the benefits of our model seem to come exclusively from modeling the local context, even for the longest documents.[6] (iii) In order to evaluate our approach, we have created oracle labels for both Pubmed and arXiv BIBREF9 , by applying a greedy oracle labeling algorithm. The two datasets annotated with extractive labels will be made public.
Our main contributions are as follows: (i) In order to capture the local context, we are the first to apply LSTM-minus to text summarization. LSTM-minus is a method for learning embeddings of text spans, which has achieved good performance in dependency parsing, in constituency parsing, as well as in discourse parsing. With respect to more traditional methods for capturing local context, which rely on hierarchical structures, LSTM-minus produces simpler models i.e. with less parameters, and therefore faster to train and less prone to overfitting. (ii) We test our method on the Pubmed and arXiv datasets and results appear to support our goal of effectively summarizing long documents. In particular, while overall we outperform the baseline and previous approaches only by a narrow margin on both datasets, the benefit of our method become much stronger as we apply it to longer documents. purpleFurthermore, in an ablation study to assess the relative contributions of the global and the local model we found that, rather surprisingly, the benefits of our model seem to come exclusively from modeling the local context, even for the longest documents.(iii) In order to evaluate our approach, we have created oracle labels for both Pubmed and arXiv, by applying a greedy oracle labeling algorithm.
|
What are the author's main contributions?
|
(i) In order to capture the local context, they are the first to apply LSTM-minus to text summarization. (ii) They test their method on the Pubmed and arXiv datasets and results appear to support their goal of effectively summarizing long documents. (iii) In order to evaluate their approach, they have created oracle labels for both Pubmed and arXiv, by applying a greedy oracle labeling algorithm.
|
1907.05664
| false
| null |
The baseline model is a deep sequence-to-sequence encoder/decoder model with attention. The encoder is a bidirectional Long-Short Term Memory(LSTM) cell BIBREF14 and the decoder a single LSTM cell with attention mechanism. The attention mechanism is computed as in BIBREF9 and we use a greedy search for decoding. We train end-to-end including the words embeddings. The embedding size used is of 128 and the hidden state size of the LSTM cells is of 254.
The baseline model is a deep sequence-to-sequence encoder/decoder model with attention. The encoder is a bidirectional Long-Short Term Memory(LSTM) cell BIBREF14 and the decoder a single LSTM cell with attention mechanism. The attention mechanism is computed as in BIBREF9 and we use a greedy search for decoding. We train end-to-end including the words embeddings. The embedding size used is of 128 and the hidden state size of the LSTM cells is of 254.
|
Which baselines did they compare?
|
The answers are shown as follows:
* The baseline model is a deep sequence-to-sequence encoder/decoder model with attention. The encoder is a bidirectional Long-Short Term Memory(LSTM) cell BIBREF14 and the decoder a single LSTM cell with attention mechanism. The attention mechanism is computed as in BIBREF9 and we use a greedy search for decoding. We train end-to-end including the words embeddings. The embedding size used is of 128 and the hidden state size of the LSTM cells is of 254.
|
null | false
| 172
|
The dataset was thoroughly evaluated, see leitner2019fine for more details. As state of the art models, Conditional Random Fields (CRFs) and bidirectional Long-Short Term Memory Networks (BiLSTMs) were tested with the two variants of annotation. For CRFs, these are: CRF-F (with features), CRF-FG (with features and gazetteers), CRF-FGL (with features, gazetteers and lookup). For BiLSTM, we used models with pre-trained word embeddings BIBREF22: BiLSTM-CRF BIBREF23, BiLSTM-CRF+ with character embeddings from BiLSTM BIBREF24, and BiLSTM-CNN-CRF with character embeddings from CNN BIBREF25. To evaluate the performance we used stratified 10-fold cross-validation. As expected, BiLSTMs perform best (see Table ). The F$_1$ score for the fine-grained classification reaches 95.46 and 95.95 for the coarse-grained one. CRFs reach up to 93.23 F$_1$ for the fine-grained classes and 93.22 F$_1$ for the coarse-grained ones. Both models perform best for judge, court and law.
To evaluate the performance we used stratified 10-fold cross-validation.
|
What was used to evaluate the performance?
|
The authors used stratified 10-fold cross-validation.
|
null | false
| null |
Dance Mums with Jennifer Ellison is a British reality television series that made its debut on Lifetime on 20 October 2014. Created by Shiver Productions, it is set in Liverpool, England, at Jelli Studios and follows the early careers of children in dance and show business, as well as the participation of their mothers. It is a spin-off of the American TV series Dance Moms.
The second series began airing on 12 October 2015, but the show was not commissioned for a third series.
|
How many series did Dance Mums air?
|
Two seasons, it was not commissioned for a third series.
|
1909.08824
| false
| null |
We first compare the perplexity of CWVAE with baseline methods. Perplexity measures the probability of model to regenerate the exact targets, which is particular suitable for evaluating the model performance on one-to-many problem BIBREF20. Further, we employ BLEU score to evaluate the accuracy of generations BIBREF21, and the number of distinct n-gram to evaluate the diversity of generations BIBREF6. The distinct is normalized to $[0, 1]$ by dividing the total number of generated tokens.
FLOAT SELECTED: Table 4: Average perplexity and BLEU score (reported in percentages) for the top 10 generations under each inference dimension of Event2Mind. The the best result for each dimension is emboldened.
FLOAT SELECTED: Table 6: Average perplexity and BLEU scores (reported in percentages) for the top 10 generations under each inference dimension of Atomic. The the best result for each dimension is emboldened.
Further, we employ BLEU score to evaluate the accuracy of generations BIBREF21, and the number of distinct n-gram to evaluate the diversity of generations BIBREF6.
FLOAT SELECTED: Table 4: Average perplexity and BLEU score (reported in percentages) for the top 10 generations under each inference dimension of Event2Mind. The the best result for each dimension is emboldened.
FLOAT SELECTED: Table 6: Average perplexity and BLEU scores (reported in percentages) for the top 10 generations under each inference dimension of Atomic. The the best result for each dimension is emboldened.
|
By how much do they improve the accuracy of inferences over state-of-the-art methods?
|
ON Event2Mind, the accuracy of proposed method is improved by absolute BLUE 2.9, 10.87, 1.79 for xIntent, xReact and oReact respectively.
On Atomic dataset, the accuracy of proposed method is improved by absolute BLUE 3.95. 4.11, 4.49 for xIntent, xReact and oReact.respectively.
|
null | false
| null |
Classify these as a synthetic or natural fiber
|
wool, polyester, cotton, silk, nylon, aramid, jute
|
|
null | false
| 197
|
This paper proposes a machine learning approach to part-of-speech tagging and named entity recognition for Greek, focusing on the extraction of morphological features and classification of tokens into a small set of classes for named entities. The architecture model that was used is introduced. The greek version of the spaCy platform was added into the source code, a feature that did not exist before our contribution, and was used for building the models. Additionally, a part of speech tagger was trained that can detect the morphology of the tokens and performs higher than the state-of-the-art results when classifying only the part of speech. For named entity recognition using spaCy, a model that extends the standard ENAMEX type (organization, location, person) was built. Certain experiments that were conducted indicate the need for flexibility in out-of-vocabulary words and there is an effort for resolving this issue. Finally, the evaluation results are discussed.
The greek version of the spaCy platform was added into the source code, a feature that did not exist before our contribution, and was used for building the models.
|
Are there any highlights in their work?
|
Yes. The greek version of the spaCy platform was added to the source code and was used for building the models.
|
null | false
| null |
An artificial general intelligence (AGI) is a hypothetical intelligent agent which can understand or learn any intellectual task that human beings or other animals can. AGI has also been defined alternatively as autonomous systems that surpass human capabilities at the majority of economically valuable work. Developing AGI is a primary goal of some artificial intelligence research and for several AI companies such as OpenAI, DeepMind, and Anthropic, as well as a common topic in science fiction and futures studies.
|
What's an AGI?
|
AGI (artificial general intelligence) refers to a hypothetical intelligent agent which can understand or learn any intellectual task that human beings or other animals can. It has also been defined as autonomous agents that could surpass human capabilities at the majority of economically valuable work.
|
null | false
| null |
Who is the best football team in history?
|
Liverpool Football Club
|
|
null | false
| null |
Identify which instrument is string or percussion: Frame drum, Citole
|
Citole is string, Frame drum is percussion.
|
|
null | false
| null |
How can I drive from South Bay to San Francisco?
|
To drive from South Bay to San Francisco you will want to take either Interstate 280 or US Route 101. 101 tends to be a more direct drive but also tends to have more traffic congestion through the day. 280 is longer and hillier but tends to have less traffic and has the added bonus of nice scenery.
|
|
null | false
| null |
Give me a bulleted list of different ways to get around a city without a car
|
The following are different ways to get around a city without a car:
1. Walk
2. Take public transportation such as the bus or subway
3. Ride a bike
4. Use Uber
5. Ride a scooter
6. Ask a friend with a car
7. Hitchhike
|
|
null | false
| null |
Think of three different colors that mens suits are often
|
Black, blue, grey
|
|
null | false
| null |
Pick a title for the following:
The purpose of coaching is to form a partnership that enables you to draw more consistently on your wisdom, positive energy, and resourcefulness in decision-making, living a full life, and achieving your personal and professional goals by expanding your perspectives, being introspective, reducing blind spots, effectively managing self-limiting beliefs and feelings that block you, and developing new skills, habits, and practices.
|
Personal and Professional Coaching Goals
|
|
null | false
| null |
Tell me which of these are NFL teams: Bengals, Pumas, Seahawks, Falcons, Roadrunners, Eagles, Coyotes, Sheep, Bears.
|
Bengals, Seahawks, Falcons, Eagles, Bears.
|
|
null | false
| null |
is customer service a great career
|
yes
|
|
null | false
| null |
The McIntosh County Seat War was a dispute in Oklahoma over the location of the McIntosh County seat that took place between 1907 and 1909. Following a pair of elections that resulted in the town of Checotah being designated as the new county seat, the people of Eufaula refused to hand over the county records. As a result, a group of heavily armed men from Chectotah attempted to seize the records, but were forced to surrender during the gunbattle that ensued. One year later, after another close election, Eufaula became the permanent county seat.
|
What was the The McIntosh County Seat War
|
The McIntosh County Seat War was a dispute in Oklahoma over the location of the McIntosh County seat that took place between 1907 and 1909. Following a pair of elections that resulted in the town of Checotah being designated as the new county seat, the people of Eufaula refused to hand over the county records.
As a result, a group of heavily armed men from Chectotah attempted to seize the records, but were forced to surrender during the gunbattle that ensued. One year later, after another close election, Eufaula became the permanent county seat.
|
1704.02686
| true
| null |
Outlier Detection. The Outlier Detection task BIBREF0 is to determine which word in a list INLINEFORM0 of INLINEFORM1 words is unrelated to the other INLINEFORM2 which were chosen to be related. For each INLINEFORM3 , one can compute its compactness score INLINEFORM4 , which is the compactness of INLINEFORM5 . INLINEFORM6 is explicitly computed as the mean similarity of all word pairs INLINEFORM7 . The predicted outlier is INLINEFORM8 , as the INLINEFORM9 related words should form a compact cluster with high mean similarity.
Sentiment analysis. We also consider sentiment analysis as described by BIBREF31 . We use the suggested Large Movie Review dataset BIBREF32 , containing 50,000 movie reviews.
Outlier Detection. The Outlier Detection task BIBREF0 is to determine which word in a list INLINEFORM0 of INLINEFORM1 words is unrelated to the other INLINEFORM2 which were chosen to be related. For each INLINEFORM3 , one can compute its compactness score INLINEFORM4 , which is the compactness of INLINEFORM5 .
Sentiment analysis. We also consider sentiment analysis as described by BIBREF31 .
|
Do they test their word embeddings on downstream tasks?
|
Yes.
|
null | false
| 98
|
Vietnamese, like many languages in continental East Asia, is an isolating language and one branch of Mon-Khmer language group. The most basic linguistic unit in Vietnamese is morpheme, similar with syllable or token in English and “hình vị” (phoneme) or “tiếng” (syllable) in Vietnamese. According to the structured rule of its, Vietnamese can have about 20,000 different syllables (tokens). However, there are about 8,000 syllables used the Vietnamese dictionaries. There are three methods to identify morphemes in Vietnamese text BIBREF10 .
Morpheme is the smallest meaningful unit of Vietnamese.
Morpheme is the basic unit of Vietnamese.
Morpheme is the smallest meaningful unit and is not used independently in the syntax factor.
In computational linguistics, morpheme is the basic unit of languages as Leonard Bloomfield mentioned for English BIBREF11 . In our research for Vietnamese, we consider the morpheme as syllable, called “tiếng” in Vietnamese (as Nguyen’s definition BIBREF12 ).
The next concept in linguistics is word which has fully grammar and meaning function in sentences. For Vietnamese, word is a single morpheme or a group of morphemes, which are fixed and have full meaning BIBREF12 . According to Nguyen, Vietnamese words are able classified into two types, (1) 1- syllable words with fully meaning and (2) n-syllables words whereas these group of tokens are fixed. Vietnamese syllable is not fully meaningful. However, it is also explained in the meaning and structure characteristics. For example, the token “kỳ” in “quốc kỳ” whereas “quốc” means national, “kỳ” means flag. Therefore, “quốc kỳ” means national flag.
Consider dictionary used for evaluating the corpus, extracting features for models, and evaluating the systems, there are many Vietnamese dictionaries, however we recommend the Vietnamese dictionary of Hoang Phe, so called Hoang Phe Dictionary. This dictionary has been built by a group of linguistical scientists at the Linguistic Institute, Vietnam. It was firstly published in 1988, reprinted and extended in 2000, 2005 and 2010. The dictionary currently has 45,757 word items with 15,901 Sino-Vietnamese word items (accounting for 34.75%) BIBREF13 .
Vietnamese, like many languages in continental East Asia, is an isolating language and one branch of Mon-Khmer language group.
|
Is Vietnamese an isolating language?
|
Yes.
|
null | false
| 360
|
Neural machine translation has recently achieved impressive results BIBREF0 , BIBREF1 , while learning from raw, sentence-aligned parallel text and using little in the way of external linguistic information. However, we hypothesize that various levels of linguistic annotation can be valuable for neural machine translation. Lemmatisation can reduce data sparseness, and allow inflectional variants of the same word to explicitly share a representation in the model. Other types of annotation, such as parts-of-speech (POS) or syntactic dependency labels, can help in disambiguation. In this paper we investigate whether linguistic information is beneficial to neural translation models, or whether their strong learning capability makes explicit linguistic features redundant.
Let us motivate the use of linguistic features using examples of actual translation errors by neural MT systems. In translation out of English, one problem is that the same surface word form may be shared between several word types, due to homonymy or word formation processes such as conversion. For instance, close can be a verb, adjective, or noun, and these different meanings often have distinct translations into other languages. Consider the following English INLINEFORM0 German example:
For the English source sentence in Example SECREF4 (our translation in Example SECREF5 ), a neural MT system (our baseline system from Section SECREF4 ) mistranslates close as a verb, and produces the German verb schließen (Example SECREF6 ), even though close is an adjective in this sentence, which has the German translation nah. Intuitively, part-of-speech annotation of the English input could disambiguate between verb, noun, and adjective meanings of close.
As a second example, consider the following German INLINEFORM0 English example:
German main clauses have a verb-second (V2) word order, whereas English word order is generally SVO. The German sentence (Example UID7 ; English reference in Example UID8 ) topicalizes the predicate gefährlich 'dangerous', putting the subject die Route 'the route' after the verb. Our baseline system (Example UID9 ) retains the original word order, which is highly unusual in English, especially for prose in the news domain. A syntactic annotation of the source sentence could support the attentional encoder-decoder in learning which words in the German source to attend (and translate) first.
We will investigate the usefulness of linguistic features for the language pair German INLINEFORM0 English, considering the following linguistic features:
The inclusion of lemmas is motivated by the hope for a better generalization over inflectional variants of the same word form. The other linguistic features are motivated by disambiguation, as discussed in our introductory examples.
We will investigate the usefulness of linguistic features for the language pair German↔English, considering the following linguistic features: The inclusion of lemmas is motivated by the hope for a better generalization over inflectional variants of the same word form.
|
What is motivated by the hope for a better generalization over inflectional variants of the same word form？
|
The inclusion of lemmas is motivated by the hope for a better generalization over inflectional variants of the same word form.
|
null | false
| null |
Why should we eat organic food?
|
Organic food doesn't mean you can eat endless amounts of it. It actually doesn't even mean it's healthy for you. It simply means no chemicals and pesticides are sprayed all over the crop. Some crops absorb this worse than others or even pull it up through what exists in the soil. These chemicals can be detrimental to our physical health but also to our mental health.
|
|
null | false
| null |
I'm putting together an exercise plan that needs to have both cardio and strength exercises, but I want to separate exercises into each category. Running, deadlifts, cycling, pushups, swimming, jump rope, lunges and squats.
|
Running, cycling, swimming and jump rope are all forms of cardio exercise. Deadlifts, pushups, lunges and squats are forms of strength exercise.
|
|
null | false
| 98
|
This research gathers the results of Vietnamese word segmentation of several methods into one table as show in Table II. It is noted that they are not evaluated on a same corpus. The purpose of the result illustration is to provide an overview of the results of current Vietnamese word segmentation systems based on their individual features. All studies mentioned in the table have accuracy around 94-97% based on their provided corpus.
This study also evaluates the Vietnamese word segmentation based on existing toolkits using the same annotated Vietnamese word segmentation corpus. There are two available toolkits to evaluate and to segment. To be neutral to both toolkits, we use the EVBNews Vietnamese corpus, a part of EVBCorpus, to evaluate Vietnamese word segmentation. The EVBNews corpus contains over 45,000 segmented Vietnamese sentences extracted from 1,000 general news articles (as shown in Table III) BIBREF16 . We used the same training set which has 1000 files and 45,531 sentences. vnTokenizer outputs 831,455 Vietnamese words and 1,206,475 tokens. JVnSegmenter outputs 840,387 words and 1,201,683. We correct tags (BIO), and compare to previous outputs, we have rate from vnTokenizer is 95.6% and from JVnsegmenter is 93.4%. The result of both vnTokenizer and JVnSegmenter testing on the EVBNews Vietnamese Corpus are provided in Table IV.
This study also evaluates the Vietnamese word segmentation based on existing toolkits using the same annotated Vietnamese word segmentation corpus. There are two available toolkits to evaluate and to segment. To be neutral to both toolkits, we use the EVBNews Vietnamese corpus, a part of EVBCorpus, to evaluate Vietnamese word segmentation.
|
What dataset is used for the evaluation of the toolkits?
|
EVBNews Vietnamese corpus.
|
null | false
| null |
How data engineers are different from software engineers?
|
Data engineers build data systems and databases while software engineers create applications, software, and other products.
|
|
null | false
| null |
What are some different things you should bring with you on a day hiking trip?
|
You should bring water, a snack, sunscreen, hiking boots, a first aid kit and a cellphone
|
|
null | false
| null |
Octavia Estelle Butler (June 22, 1947 – February 24, 2006) was an American science fiction author and a multiple recipient of the Hugo and Nebula awards. In 1995, Butler became the first science-fiction writer to receive a MacArthur Fellowship. Born in Pasadena, California, Butler was raised by her widowed mother. Extremely shy as a child, Butler found an outlet at the library reading fantasy, and in writing. She began writing science fiction as a teenager. She attended community college during the Black Power movement. While participating in a local writer's workshop, she was encouraged to attend the Clarion Workshop, then held in Pennsylvania, which focused on science fiction.
|
Who is Octavia Butler?
|
Octavia Butler was a black Hugo and Nebula awards winning science fiction author from Pasadena, California.
|
null | false
| 394
|
In this paper, we describe our approach and present the results for our participation in the task 1, i.e. multilingual information extraction, of the CLEF eHealth 2018 challenge BIBREF0 . More precisely, this task consists in automatically coding death certificates using the International Classification of Diseases, 10th revision (ICD-10) BIBREF1 .
We addressed the challenge by matching ICD-10 terminology entries to text phrases in death certificates. Matching text phrases to medical concepts automatically is important to facilitate tasks such as search, classification or organization of biomedical textual contents BIBREF2 . Many concept recognition systems already exist BIBREF2 , BIBREF3 . They use different approaches and some of them are open source. We developed a general purpose biomedical semantic annotation tool for our own needs. The algorithm was initially implemented to detect drugs in a social media corpora as part of the Drugs-Safe project BIBREF4 . We adapted the algorithm for the ICD-10 coding task. The main motivation in participating in the challenge was to evaluate and compare our system with others on a shared task.
The algorithm was initially implemented to detect drugs in a social media corpora as part of the Drugs-Safe project.
|
What was the algorithm initially used to?
|
It was initially used to detect drugs in a social media corpora as part of the Drugs-Safe project
|
1910.01363
| false
| null |
In order to get high precision predictions for unlabeled tweets, we choose the probability thresholds for predicting a pro-Russian or pro-Ukrainian tweet such that the classifier would achieve 80% precision on the test splits (recall at this precision level is 23%). Table TABREF38 shows the amount of polarized edges we can predict at this precision level. Upon manual inspection, we however find that the quality of predictions is lower than estimated. Hence, we manually re-annotate the pro-Russian and pro-Ukrainian predictions according to the official annotation guidelines used by BIBREF4. This way, we can label 77 new pro-Russian edges by looking at 415 tweets, which means that 19% of the candidates are hits. For the pro-Ukrainian class, we can label 110 new edges by looking at 611 tweets (18% hits). Hence even though the quality of the classifier predictions is too low to be integrated into the network analysis right away, the classifier drastically facilitates the annotation process for human annotators compared to annotating unfiltered tweets (from the original labels we infer that for unfiltered tweets, only 6% are hits for the pro-Russian class, and 11% for the pro-Ukrainian class).
This way, we can label 77 new pro-Russian edges by looking at 415 tweets, which means that 19% of the candidates are hits. For the pro-Ukrainian class, we can label 110 new edges by looking at 611 tweets (18% hits). Hence even though the quality of the classifier predictions is too low to be integrated into the network analysis right away, the classifier drastically facilitates the annotation process for human annotators compared to annotating unfiltered tweets (from the original labels we infer that for unfiltered tweets, only 6% are hits for the pro-Russian class, and 11% for the pro-Ukrainian class).
|
How can the classifier facilitate the annotation task for human annotators?
|
The answers are shown as follows:
* quality of the classifier predictions is too low to be integrated into the network analysis right away, the classifier drastically facilitates the annotation process for human annotators compared to annotating unfiltered tweets
|
null | false
| null |
The Smuggler Cove Shell Midden (Smithsonian trinomial: 35TI46) is an archeological site located in Oswald West State Park near Neahkahnie Beach, Oregon, United States. First documented by archeologists in 1976, the midden has been found to contain remains of mussels, barnacles, and chiton to a depth of approximately 30 centimeters (12 in). Radiocarbon dating of a single sample of shell debris indicates that the Smuggler Cove campsite was occupied around 1660 CE, approximately the same date as the supposed wreck of a European ship at nearby Nehalem Spit, suggesting the site may preserve information from both before and after first contact between local people and Europeans. The site has been heavily damaged by construction of recreational facilities and coastal erosion, but a significant portion remains with potential to contribute to future research. Data generated from the site may help answer questions related to environmental change in the Oregon Coast region, settlement and subsistence patterns, emergence of ethnographic patterns among coastal people, the change in cultural patterns from before to after contact with European Americans, and other topics.
|
Where is Smuggler Cove Shell Midden located?
|
Smuggler Cove shell Midden is located in Oswald West State Park near Neahkahnie Beach in Oregon, United States.
|
null | false
| null |
Why zurich is so beautiful
|
Because of it's lake, The most beautiful place in Zurich
|
|
null | false
| null |
Tardigrades are the first known animal to survive after exposure to outer space. In September 2007, dehydrated tardigrades were taken into low Earth orbit on the FOTON-M3 mission carrying the BIOPAN astrobiology payload. For 10 days, groups of tardigrades, some of them previously dehydrated, some of them not, were exposed to the hard vacuum of outer space, or vacuum and solar UV radiation. Back on Earth, more than 68% of the subjects protected from solar UV radiation were reanimated within 30 minutes following rehydration, although subsequent mortality was high; many of these produced viable embryos. In contrast, hydrated samples exposed to the combined effect of vacuum and full solar UV radiation had significantly reduced survival, with only three subjects of Milnesium tardigradum surviving. Also, it was found that the space vacuum did not have a significant effect on egg-laying in either R. coronifer or M. tardigradum. However, M. tardigradum exposed to UV radiation had a lower egg laying rate. In May 2011, Italian scientists sent tardigrades on board the International Space Station along with extremophiles on STS-134, the final flight of Space Shuttle Endeavour. Their conclusion was that microgravity and cosmic radiation "did not significantly affect survival of tardigrades in flight, and stated that tardigrades represent a useful animal for space research." In November 2011, they were among the organisms to be sent by the U.S.-based Planetary Society on the Russian Fobos-Grunt mission's Living Interplanetary Flight Experiment to Phobos; however, the launch failed. In August 2019, scientists reported that a capsule containing tardigrades in a cryptobiotic state may have survived for a while on the Moon after the April 2019 crash landing of Beresheet, a failed Israeli lunar lander, but in May 2021 it was reported that they were unlikely to have survived the impact.
|
Based on this paragraph on Tardigrades when did we first send them to space?
|
The FOTON-M3 mission in September of 2007 is where Tardigrades were exposed to the hard vacuum of outer space for the first time.
|
null | false
| null |
Are Grizzly bears and Brown bears the same?
|
Grizzly bears are actually a subspecies of brown bears (Ursus arctos horribilis). The difference is regional: Bears found inland are referred to as "grizzlies", while bears found on the coast are known as brown bears.
|
|
null | false
| null |
Qingdao Qianwan Container Terminal is a port of China, located at Qianwan (literally The Front Bay), Qingdao, Shandong province. Qingdao Port ranked 10th in 2008 world's busiest container ports with 10,320 thousands TEU, current projects were expected to raise the capacity to 15,000 thousands or more.
|
Tell me about the Qingdao Qianwan Container Terminal
|
The Qingdao Qianwan Container Terminal is a port of China, located at Qianwan (literally The Front Bay), Qingdao, Shandong province. Qingdao Port ranked 10th in 2008 world's busiest container ports with 10,320 thousands TEU, current projects were expected to raise the capacity to 15,000 thousands or more.
|
null | false
| null |
What are the top 3 largest countries by territory?
|
Russia, Canada and China
|
|
null | false
| null |
You are a young citizen of a developing country. Write a letter in less than 150 words to a newspaper, highlighting your concern over incorrect disposal of biomedical waste and to spread awareness of the same.
|
The Covid19 pandemic has changed the lives of all human beings, and has driven massive consumption of personal protective equipment worldwide. People have adapted to wearing face masks as a 'new normal'. At the same time, the disposal of biomedical waste is a major concern, especially in developing countries with limited resources. Biomedical waste includes used Rapid Antigen Tests, PCR Tests, used injections, and used masks. We have observed that people who use the test kits and masks at home, often fail to dispose them off properly. This is counter productive, as it poses the risk of aiding the virus' transmission, and can potentially be life threatening to others. Hospitals, too, produce a huge amount of waste and it is imperative they dispose it off properly following all norms.
If not scientifically handled and disposed, it could cause serious risk not just to human health, but also pose a serious threat to terrestrial and marine ecosystems.
Thank you for highlighting my concerns to your readership, and I request you to keep highlighting these concerns to your audience regularly via your articles.
|
|
null | false
| null |
What kind of Toys should be prepare for a new puppy
|
• Puppy’s teeth can be strong, and they can easily destroy soft
toys. They can also accidentally swallow part of the toys. Do
monitor the puppy when they are playing with toys to
prevent accidental choking.
• As the puppy is teething, they will need bite toys to chew
on.
So prepare the toy for the puppy base on the above consideration.
|
|
null | false
| null |
On Wednesday, September 25, 1974, the race started. It began with a run of a 4.8 km (3 mi) loop, followed by biking twice around Fiesta Island for a total of 8.0 km (5 mi). Entrants would then get off the bikes, take their shoes off and run into the water to swim to the mainland. That was followed by running in bare feet, then swimming again along the bay, then one last swim up to the entrance of Fiesta Island, and a final crawl up a steep dirt bank to finish. Most participants were not skilled swimmers, so Johnstone recruited his 13-year-old son to float on his surfboard and act as lifeguard. Some participants took longer than expected, and it began to get dark as they finished their swims. Shanahan recalls they pulled up a few cars and turned on the headlights so the athletes could see. Johnstone and Shanahan were surprised by the large number of entrants (46), mainly coming from local running clubs. Two notable entrants, Judy and John Collins, would four years later found the event which brought international attention to the new sport: the Hawaii Ironman
|
Given this paragraph about triathlon, who founded ironman
|
Ironman was founded by Judy and John Collins four years after the first known modern day triathlon on Wednesday, September 25th, 1974.
|
null | false
| 158
|
In this paper, we introduce Query-based Attention CNN(QACNN) for Text Similarity Map, an end-to-end neural network for question answering. This network is composed of compare mechanism, two-staged CNN architecture with attention mechanism, and a prediction layer. First, the compare mechanism compares between the given passage, query, and multiple answer choices to build similarity maps. Then, the two-staged CNN architecture extracts features through word-level and sentence-level. At the same time, attention mechanism helps CNN focus more on the important part of the passage based on the query information. Finally, the prediction layer find out the most possible answer choice. We conduct this model on the MovieQA dataset using Plot Synopses only, and achieve 79.99% accuracy which is the state of the art on the dataset.
Finally, the prediction layer find out the most possible answer choice.
|
What's the role of the prediction layer?
|
To find out the most possible answer choice.
|
null | false
| null |
Which of these are dog and which are birds: Phoenix, Siberian Husky, Alaskan Malamute, Hawk, Border Collie, Labrador, Araucana
|
Birds: Phoenix, Siberian Husky, Alaskan Malamute, Border Collie, Labrador
Dogs: Siberian Husky, Alaskan Malamute, Border Collie, Labrador
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.