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 |
Hoober Stand is a 30-metre-high (98 ft) tower and Grade II* listed building on a ridge in Wentworth, South Yorkshire in northern England. It was designed by Henry Flitcroft for the Whig aristocrat Thomas Watson-Wentworth, Earl of Malton (later the 1st Marquess of Rockingham) to commemorate the quashing of the 1745 Jacobite rebellion. It lies close to his country seat Wentworth Woodhouse. Its site is approximately 157 metres (515 ft) above sea level and from the top there are long-distance views on a clear day. Hoober Stand is one of several follies in and around Wentworth Woodhouse park; the others include Needle's Eye and Keppel's Column. Sidney Oldall Addy, the Sheffield author calls the structure Woburn Stand in his 1888 book, A glossary of words used in the neighbourhood of Sheffield.
|
Where is Hoober Stand located?
|
Wentworth, South Yorkshire in northern England
|
null | false
| 56
|
Earlier studies on stock market prediction are based on the historical stock prices. Later studies have debunked the approach of predicting stock market movements using historical prices. Stock market prices are largely fluctuating. The efficient market hypothesis (EMH) states that financial market movements depend on news, current events and product releases and all these factors will have a significant impact on a company's stock value BIBREF0 . Because of the lying unpredictability in news and current events, stock market prices follow a random walk pattern and cannot be predicted with more than 50% accuracy BIBREF1 .
With the advent of social media, the information about public feelings has become abundant. Social media is transforming like a perfect platform to share public emotions about any topic and has a significant impact on overall public opinion. Twitter, a social media platform, has received a lot of attention from researchers in the recent times. Twitter is a micro-blogging application that allows users to follow and comment other users thoughts or share their opinions in real time BIBREF2 . More than million users post over 140 million tweets every day. This situation makes Twitter like a corpus with valuable data for researchers BIBREF3 .Each tweet is of 140 characters long and speaks public opinion on a topic concisely. The information exploited from tweets are very useful for making predictions BIBREF4 .
In this paper, we contribute to the field of sentiment analysis of twitter data. Sentiment classification is the task of judging opinion in a piece of text as positive, negative or neutral.
There are many studies involving twitter as a major source for public-opinion analysis. Asur and Huberman BIBREF5 have predicted box office collections for a movie prior to its release based on public sentiment related to movies, as expressed on Twitter. Google flu trends are being widely studied along with twitter for early prediction of disease outbreaks. Eiji et al. BIBREF6 have studied the twitter data for catching the flu outbreaks. Ruiz et al. BIBREF7 have used time-constrained graphs to study the problem of correlating the Twitter micro-blogging activity with changes in stock prices and trading volumes. Bordino et al. BIBREF8 have shown that trading volumes of stocks traded in NASDAQ-100 are correlated with their query volumes (i.e., the number of users requests submitted to search engines on the Internet). Gilbert and Karahalios BIBREF9 have found out that increases in expressions of anxiety, worry and fear in weblogs predict downward pressure on the S&P 500 index. Bollen BIBREF10 showed that public mood analyzed through twitter feeds is well correlated with Dow Jones Industrial Average (DJIA). All these studies showcased twitter as a valuable source and a powerful tool for conducting studies and making predictions.
Rest of the paper is organized as follows. Section 2 describes the related works and Section 3 discusses the data portion demonstrating the data collection and pre-processing part. In Section 4 we discuss the sentiment analysis part in our work followed by Section 5 which examines the correlation part of extracted sentiment with stocks. In Section 6 we present the results, accuracy and precision of our sentiment analyzer followed by the accuracy of correlation analyzer. In Section 7 we present our conclusions and Section 8 deals with our future work plan.
Sentiment classification is the task of judging opinion in a piece of text as positive, negative or neutral.
|
What is sentiment classification?
|
It is the task of judging opinion in a piece of text as positive, negative or neutral.
|
null | false
| null |
Are you going for a walk today?
|
After these last rainy days, I'm definitely going for a walk today! With this sunny weather, staying at home would be a shame!
|
|
null | false
| null |
What is a cat?
|
A cat is a cute, cuddly animal with whiskers, pointed ears, four legs and paws, and a tail. Cats are companion animals to humans. Cats can come in many different varieties, such as hairless, short-haired, medium-haired, and very fluffy; they also come in many colors, such as black, white, grey, brown, orange, and a combination of multiple colors.
|
|
null | false
| null |
Which Gene in the body is responsible for a persons height
|
A variant called HMGA2 is responsible for persons height
|
|
1604.05781
| false
| null |
Causal documents were chosen to contain one occurrence only of the exact unigrams: `caused', `causing', or `causes'. The word `cause' was not included due to its use as a popular contraction for `because'. One `cause-word' per document restricted the analysis to single relationships between two relata. Documents that contain bidirectional words (`associate', `relate', `connect', `correlate', and any of their stems) were also not selected for analysis. This is because our focus is on causality, an inherently one-sided relationship between two objects. We also did not consider additional synonyms of these cause words, although that could be pursued for future work. Control documents were also selected. These documents did not contain any of `caused', `causing', or `causes', nor any bidirectional words, and are further matched temporally to obtain the same number of control documents as causal documents in each fifteen-minute period during 2013. Control documents were otherwise selected randomly; causal synonyms may be present. The end result of this procedure identified 965,560 causal and 965,560 control documents. Each of the three “cause-words”, `caused', `causes', and `causing' appeared in 38.2%, 35.0%, and 26.8% of causal documents, respectively.
Causal documents were chosen to contain one occurrence only of the exact unigrams: `caused', `causing', or `causes'.
|
How do they extract causality from text?
|
They identify documents that contain the unigrams 'caused', 'causing', or 'causes'
|
null | false
| 235
|
Author profiling is the characterization of an author through some key attributes such as gender, age, and language. It's an indispensable task especially in security, forensics, and marketing. Recently, social media has become a great data source for the potential learning approaches. Furthermore, gender prediction has been a popular profiling task.
The traditional approach to gender prediction problem is extracting a useful set of hand-crafted features and then feeding them into a standard classification algorithm. In their study, BIBREF0 work with the style-based features of message length, stop word usage, frequency of smiley etc. and use different classifiers such as k-nearest neighbor, naive bayes, covering rules, and backpropagation to predict gender on chat messages. Similarly, BIBREF1 select some hand-crafted features and feed them into various classifiers.
Most of the work on gender prediction rely on n-gram features BIBREF2. BIBREF3 give Latent Semantic Analysis (LSA)-reduced forms of word and character n-grams into Support Vector Machine (SVM) and achieve state-of-the-art performance. Apart from exploiting n-gram frequencies, there are studies BIBREF4, BIBREF5, BIBREF6 to extract cross-lingual features to determine gender from tweets. Some other work BIBREF4, BIBREF7 exploit user metadata besides using just tweets.
Recently, neural network-based models have been proposed to solve this problem. Rather than explicitly extracting features, the aim is to develop an architecture that implicitly learns. In author profiling, both style and content-based features were proved useful BIBREF8 and neural networks are able to capture both syntactic and semantic regularities. In general, syntactic information is drawn from the local context. On the other hand, semantic information is often captured with larger window sizes. Thus, CNNs are preferred to obtain style-based features while RNNs are the methods of choice for addressing content-based features BIBREF9. In literature, CNN BIBREF10 or RNN BIBREF11, BIBREF12, BIBREF13 is used on this task. BIBREF11 obtain state-of-the-art performance among neural methods by proposing a model architecture where they process text through RNN with GRU cells. Also, the presence of an attention layer is shown to boost the performance of neural methods BIBREF11, BIBREF10.
In this work, we propose a model that relies on RNN with attention mechanism (RNNwA). A bidirectional RNN with attention mechanism both on word level and tweet level is trained with word embeddings. The final representation of the user is fed to a fully connected layer for prediction. Since combining some hand-crafted features with a learned linear layer has shown to perform well in complex tasks like Semantic Role Labeling (SRL) BIBREF14, an improved version of the model (RNNwA + n-gram) is also tested with hand-crafted features. In the improved version, LSA-reduced n-gram features are concatenated with the neural representation of the user. Then the result is fed into a fully-connected layer to make prediction. Models are tested in three languages; English, Spanish, and Arabic, and the improved version achieves state-of-the-art accuracy on English, and competitive results on Spanish and Arabic corpus.
There are many datasets created for this task BIBREF15, BIBREF16. In this work, we have used the dataset and benchmarks provided by the PAN 2018 shared task on author profiling BIBREF15. As the dataset contains a constant number of 100 tweets per user, accuracy tests are performed both on user and tweet level (tweet-level predictions are made by removing the user-level attention). Tweet-level accuracy tests show interesting results during hyperparameter optimization. When the tweet-level predictions are averaged to produce user-level predictions, it is seen that the hyperparameters that gave the best results in terms of tweet-level accuracy, performs worse in user-level accuracy. The better user-level models, with different hyperparameters, that gave the highest user-level accuracy are observed to slightly overfit on tweet-level. It leads us to believe that the overfitting in the tweet-level predictions in best user-level models acts similar to an attention mechanism by over-emphasizing some distinctive tweets and ignoring the rest.
The final representation of the user is fed to a fully connected layer for prediction.
|
Where does the final representation of the user is fed to?
|
To a fully connected layer for prediction.
|
null | false
| null |
What is Big Data?
|
Big data is a term used to describe sets of data that are too large and complex to be analyzed and understood using traditional methods and tools.
|
|
null | false
| null |
Tell me whether these are literary genres or titles of books: Non-fiction, 1984, Mystery, The lost girls of paris, Fiction, Narrative, The one and only Ruby, Mary Poppins, Science fiction.
|
Sure, literary genres include: Non-fiction, Mystery, Fiction, Narrative, and Science fiction. Titles of books from this list are: 1984, The Lost girls of Paris, The One and Only Ruby, Marry Poppins.
|
|
null | false
| 69
|
Despite its usefulness, linked entities extracted from ELS's have issues because of low precision rates BIBREF11 and design challenges in training datasets BIBREF12 . These issues can be summarized into two parts: ambiguity and coarseness.
First, the extracted entities may be ambiguous. In the example, the entity “South Korean” is ambiguous because it can refer to both the South Korean person and the South Korean language, among others. In our experimental data, we extract (1) the top 100 entities based on frequency, and (2) the entities extracted from 100 randomly selected texts, and check whether they have disambiguation pages in Wikipedia or not. We discover that $71.0\%$ of the top 100 entities and $53.6\%$ of the entities picked at random have disambiguation pages, which shows that most entities are prone to ambiguity problems.
Second, the linked entities may also be too common to be considered an entity. This may introduce errors and irrelevance to the summary. In the example, “Wednesday” is erroneous because it is wrongly linked to the entity “Wednesday Night Baseball”. Also, “swap” is irrelevant because although it is linked correctly to the entity “Trade (Sports)”, it is too common and irrelevant when generating the summaries. In our experimental data, we randomly select 100 data instances and tag the correctness and relevance of extracted entities into one of four labels: A: correct and relevant, B: correct and somewhat relevant, C: correct but irrelevant, and D: incorrect. Results show that $29.4\%$ , $13.7\%$ , $30.0\%$ , and $26.9\%$ are tagged with A, B, C, and D, respectively, which shows that there is a large amount of incorrect and irrelevant entities.
Despite its usefulness, linked entities extracted from ELS’s have issues because of low precision rates (Hasibi et al., 2016) and design challenges in training datasets (Ling et al., 2015). These issues can be summarized into two parts: ambiguity and coarseness.
|
What issues do the linked entities extracted from ELS's have?
|
These issues can be summarized into two parts: ambiguity and coarseness.
|
null | false
| null |
Identify which animal species is alive or extinct: Carolina Parakeet, Orangutan
|
Orangutan is alive, Carolina Parakeet is extinct.
|
|
null | false
| null |
Lighthouse Point, Bahamas, or simply Lighthouse Point, is a private peninsula in The Bahamas which serves as an exclusive port for the Disney Cruise Line ships. It is located in the south-eastern region of Bannerman Town, Eleuthera. In March 2019, The Walt Disney Company purchased the peninsula from the Bahamian government, giving the company control over the area.
|
Where is the Lighthouse Point, Bahamas
|
The Lighthouse Point, Bahamas, or simply Lighthouse Point, is a private peninsula in the Bahamas which serves as an exclusive port for the Disney Cruise Line ships. It is located in the south-eastern region of Bannerman Town, Eleuthera.
|
null | false
| 2
|
In this section, we show the model performances of our proposed crowdsourcing learning system (ALCrowd), and meanwhile compare it with the other systems mentioned above. Table 2 shows the experimental results on the DL-PS datasets and Table 3 shows the experiment results on the EC-MT and EC-UQ datasets, respectively.
The results of CRF and LSTM-CRF mean that the crowd annotation is an alternative solution with low cost for labeling data that could be used for training a NER system even there are some inconsistencies. Compared with CRF, LSTM-CRF achieves much better performances on all the three data, showing +6.12 F1 improvement on DL-PS, +4.51 on EC-MT, and +9.19 on EC-UQ. This indicates that LSTM-CRF is a very strong baseline system, demonstrating the effectiveness of neural network.
Interestingly, when compared with CRF and LSTM-CRF, CRF-VT and LSTM-CRF-VT trained on the voted version perform worse in the DL-PS dataset. This trend is also mentioned in BIBREF4 nguyen2017aggregating. This fact shows that the majority voting method might be unsuitable for our task. There are two possible reasons accounting for the observation. On the one hand, simple character-level voting based on three annotations for each sentence may be still not enough. In the DL-PS dataset, even with only two predefined entity types, one character can have nine NE labels. Thus the majority-voting may be incapable of handling some cases. While the cost by adding more annotations for each sentence would be greatly increased. On the other hand, the lost information produced by majority-voting may be important, at least the ambiguous annotations denote that the input sentence is difficult for NER. The normal CRF and LSTM-CRF models without discard any annotations can differentiate these difficult contexts through learning.
Three crowd-annotation learning systems provide better performances than their counterpart systems, (CRF-MA VS CRF) and (LSTM-Crowd/ALCrowd VS LSTM-CRF). Compared with the strong baseline LSTM-CRF, ALCrowd shows its advantage with +1.08 F1 improvements on DL-PS, +1.24 on EC-MT, and +2.38 on EC-UQ, respectively. This indicates that adding the crowd-annotation learning is quite useful for building NER systems. In addition, ALCrowd also outperforms LSTM-Crowd on all the datasets consistently, demonstrating the high effectiveness of ALCrowd in extracting worker independent features. Among all the systems, ALCrowd performs the best, and significantly better than all the other models (the p-value is below $10^{-5}$ by using t-test). The results indicate that with the help of adversarial training, our system can learn a better feature representation from crowd annotation.
Among all the systems, ALCrowd performs the best, and significantly better than all the other models (the p-value is below 10−5 by using t-test).
|
Which systems perform the best?
|
Among all the systems, ALCrowd performs the best, and significantly better than all the other models (the p-value is below 10−5 by using t-test).
|
null | false
| null |
What are the names of some of the cities in the Boston Metro West area?
|
Boston Metro West cities include Newton, Natick, Framingham, Wellesley, Dedham, and Needham, among others.
|
|
null | false
| null |
The National Gallery of Ireland (Irish: Gailearaí Náisiúnta na hÉireann) houses the national collection of Irish and European art. It is located in the centre of Dublin with one entrance on Merrion Square, beside Leinster House, and another on Clare Street. It was founded in 1854 and opened its doors ten years later. The gallery has an extensive, representative collection of Irish paintings and is also notable for its Italian Baroque and Dutch masters painting. The current director is Caroline Campbell.
|
Based on the provided text, In what year did the National Gallery of Ireland open its doors to the public and what kind of art is it known for?
|
The National Gallery of Ireland opened in 1864 and is known for its Irish, Italian Baroque, and Dutch masters paintings
|
null | false
| null |
Plan a weekend long vacation to Madrid, and mention multiple specific places you would visit and how you would travel in the city
|
On Saturday we would start off by visiting the Reina Sofia museum, which we would get to by taking the metro system. We would then continue to the Prado art museum in the afternoon, which we would also get to by metro and walking from the Reina Sofia. We'd then grab a late lunch at a nearby cafe, before walking over to the nearby Parque Retiro, which is one of the main parks in Madrid. We'd spend the afternoon there and take out a boat for a row on the lake, and then in the evening we'd grab empanadas for dinner from a bakery. The next day, we'd start off the day with a casual coffee and slow morning breakfast. We'd then check out the botanical gardens, and then the National Anthropological Museum in the afternoon. Finally, we'd finish our weekend with a visit to the Madrid Opera house.
|
|
null | false
| 517
|
When a single edge device only has limited number of target data, it is hard to learn the fine-grained representations of the target data even for the large model. As a result, the compact model trained by Co-KD still suffers low inference accuracy on the target data. To tackle this challenging, Co-KD can be extended to unify the target data of the multiple edge devices for training a global compact model.****Two group of experiments are conducted in the case of multiple edge devices. First, the target data on the multiple edge devices come from the same target domain. Second, the target data on the multiple edge devices come from different target domains. More specifically, in the second experiments, the edge devices are divided into two groups. The devices from the same group have the target data from the same domain, while the target domains of the two groups are different.
|
The federated learning setting is realistic because it can extend to scenarios with many clients. However, this paper mainly considers one client. The multiple clients in the appendix all come from one target domain, but the realistic setting is that different clients come from different domains. Do the author consider it?
|
Indeed we mainly consider one client in the main text. The reason is as follows. If the clients are assumed to come from the same target domain, our method is applicable to either a single client or multiple clients. For the clarity of presentation, in the main text our method is presented under the setup of a single client. Furthermore, the scenario of multiple clients is presented in the appendix. For the scenario where multiple clients come from different target domains, we will conduct additional experiments to evaluate the performance of our scheme. It should be noted that our scheme is not specifically designed for different target domains, so the model may not perform equally well on all the target domains. Tackling the issue of different target domains is an interesting topic for further study.
|
null | false
| null |
Ihor Lapin (born 28 May 1969 in Ivanychi) is a Ukrainian lawyer and politician who is a former (2014-2019) Member of the Parliament of Ukraine.
During the War in Donbass (eastern Ukraine) he was commander of the 2nd assault squadron "West" of the 24th Battalion of Territorial Defense "Aidar" of the Armed Forces of Ukraine, formed mainly from volunteers from Volyn Region.
In the Russian-Ukrainian war, he was the commander of the Special Forces Battalion of the First Separate Special Forces Brigade, named after Ivan Bohun.
Before the war, he worked as a lawyer, and was a member of the Bar Qualification-Disciplinary Commission of Volyn Region.
|
Given a reference text about Ihor Lapin, tell me what he did before the war.
|
Before the war, Ihor Lapin was a lawyer and a member of the Bar Qualification-Disciplinary Commission.
|
null | false
| null |
How many atomic bombs have been used in combat in human history?
|
Two atomic bombs have been deployed against populations in human history
|
|
null | false
| null |
Identify which instrument is string or woodwind: Harpsichord, Quena
|
Quena is woodwind, Harpsichord is string.
|
|
null | false
| null |
What are some interesting facts about tigers?
|
1. Tiger stripes are unique.
2. Tigers can use their ears to communicate.
3. India has over half the world's population of tigers.
4. Only one in ten hunts are successful.
|
|
null | false
| null |
Classify each of the following as either a city or a country: Seoul, Shanghai, Sao Paolo, Singapore, Sweden.
|
Seoul is the largest city in South Korea and also the capital. Shanghai is the most populous city in China and is also known as "The Pearl of Asia." Sao Paolo is a city in the Brazilian state of the same name, Sao Paolo and is named after an Apostle, Saint Paul of Tarsus. Singapore is simultaneously a city, state, and country. Sweden is a country which has produced many world-renowned pop singers and bands including Abba.
|
|
null | false
| null |
What are some ways to care for a new tree
|
There are many ways to care for a young tree to ensure its future health and success. Some ways you may want to care for a young tree include:
- Frequent watering. The amount of water and frequency will vary depending on the tree and climate but frequent, deep watering is essential for early development.
- Nutrients. Fertilizers may aid in the early stages. Be sure to read the manufacturers label and instructions before application.
- Mulch. Applying mulch to the base of the tree aids in weeds suppression and moisture retention.
- Remove nearby weeds/plants. Weeds and plants can provide competition to your tree for important resources. Remove any near the root system of your tree.
These are just some ways you can ensure a healthy start to your tree!
|
|
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.
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.
|
What do the authors train in this paper?
|
The authors train both Support Vector Machines (SVM) and Feed-forward Neural Networks (FNN) obtaining state-of-the-art results.
|
null | false
| 178
|
A core problem in artificial intelligence is to capture, in machine-usable form, the collection of information that an ordinary person would have, known as commonsense knowledge. For example, a machine should know that a room may have a door, and that when a person enters a room, it is generally through a door. This background knowledge is crucial for solving many difficult, ambiguous natural language problems in coreference resolution and question answering, as well as the creation of other reasoning machines.
More than just curating a static collection of facts, we would like commonsense knowledge to be represented in a way that lends itself to machine reasoning and inference of missing information. We concern ourselves in this paper with the problem of learning commonsense knowledge representations.
In machine learning settings, knowledge is usually represented as a hypergraph of triplets such as Freebase BIBREF1 , WordNet BIBREF2 , and ConceptNet BIBREF3 . In these knowledge graphs, nodes represent entities or terms $t$ , and hyperedges are relations $R$ between these entities or terms, with each fact in the knowledge graph represented as a triplet $<t_1, R, t_2>$ . Researchers have developed many models for knowledge representation and learning in this setting BIBREF4 , BIBREF5 , BIBREF6 , BIBREF7 , BIBREF8 , under the umbrella of knowledge graph completion. However, none of these naturally lend themselves to traditional methods of logical reasoning such as transitivity and negation.
While a knowledge graph completion model can represent relations such as Is-A and entailment, there is no mechanism to ensure that its predictions are internally consistent. For example, if we know that a dog is a mammal, and a pit bull is a dog, we would like the model to also predict that a pit bull is a mammal. These transitive entailment relations describe ontologies of hierarchical data, a key component of commonsense knowledge which we focus on in this work.
Recently, a thread of research on representation learning has aimed to create embedding spaces that automatically enforce consistency in these predictions using the intrinsic geometry of the embedding space BIBREF9 , BIBREF0 , BIBREF10 . In these models, the inferred embedding space creates a globally consistent structured prediction of the ontology, rather than the local relation predictions of previous models.
We focus on the order-embedding model BIBREF0 which was proposed for general hierarchical prediction including multimodal problems such as image captioning. While the original work included results on ontology prediction on WordNet, we focus exclusively on the model's application to commonsense knowledge, with its unique characteristics including complex ordering structure, compositional, multi-word entities, and the wealth of commonsense knowledge to be found in large-scale unstructured text data.
We propose two extensions to the order embedding model. The first augments hierarchical supervision from existing ontologies with non-hierarchical knowledge in the form of raw text. We find incorporating unstructured text brings accuracy from 92.0 to 93.0 on a commonsense dataset containing Is-A relations from ConceptNet and Microsoft Concept Graph (MCG), with larger relative gains from smaller amounts of labeled data.
The second extension uses the complex partial-order structure of real-world ontologies to find long-distance triplet constraints among embeddings which are poorly enforced by the standard pairwise training method. By adding our additional triplet constraints to the baseline order-embedding model, we find performance improves from 90.6 to 91.3 accuracy on the WordNet ontology dataset.
We find that order embeddings' ease of extension, both by incorporating non-ordered data, and additional training constraints derived from the structure of the problem, makes it a promising avenue for the development of further algorithms for automatic learning and jointly consistent prediction of ontologies.
We propose two extensions to the order embedding model. The first augments hierarchical supervision from existing ontologies with non-hierarchical knowledge in the form of raw text. We find incorporating unstructured text brings accuracy from 92.0 to 93.0 on a commonsense dataset containing Is-A relations from ConceptNet and Microsoft Concept Graph (MCG), with larger relative gains from smaller amounts of labeled data. The second extension uses the complex partial-order structure of real-world ontologies to find long-distance triplet constraints among embeddings which are poorly enforced by the standard pairwise training method.
|
What are the two extensions they propose to the order-embedding model?
|
The first augments hierarchical supervision from existing ontologies with non-hierarchical knowledge in the form of raw text. The second extends the complex partial-order structure of real-world ontologies to find long-distance triplet constraints among embeddings which are poorly enforced by the standard pairwise training method.
|
null | false
| null |
Tell me which following countries are in Asia: United States, Canada, Britain, France, Australia, Korea, China, Egypt, Israel, Japan, Brazil
|
Given your list, only Korea, China, and Japan are countries in Asia.
|
|
null | false
| null |
What family was well known for their power in Renaissance Italy?
|
The Medici family was a famously influential presence in Renaissance Italy
|
|
null | false
| null |
Think of some family rules to promote a healthy family relationship
|
Use Kind words
Be grateful
Think of others before yourself
Work hard
Be Honest
Be thankful
Keep your promises
Choose to be happy
Always say 'I Love You'
Be kind
Encourage each other
|
|
1705.07368
| false
| null |
I also performed several case studies. I obtained document embeddings, in the same latent space as the topic embeddings, by summing the posterior mean vectors INLINEFORM0 for each token, and visualized them in two dimensions using INLINEFORM1 -SNE BIBREF24 (all vectors were normalized to unit length). The state of the Union addresses (Figure FIGREF27 ) are embedded almost linearly by year, with a major jump around the New Deal (1930s), and are well separated by party at any given time period. The embedded topics (gray) allow us to interpret the space. The George W. Bush addresses are embedded near a “war on terror” topic (“weapons, war...”), and the Barack Obama addresses are embedded near a “stimulus” topic (“people, work...”).
I also performed several case studies. I obtained document embeddings, in the same latent space as the topic embeddings, by summing the posterior mean vectors INLINEFORM0 for each token, and visualized them in two dimensions using INLINEFORM1 -SNE BIBREF24 (all vectors were normalized to unit length). The state of the Union addresses (Figure FIGREF27 ) are embedded almost linearly by year, with a major jump around the New Deal (1930s), and are well separated by party at any given time period.
|
What is an example of a computational social science NLP task?
|
Visualization of State of the union addresses
|
null | false
| null |
What does the "E" stand for in Chuck E. Cheese?
|
The "E" stands for "Entertainment"
|
|
null | false
| 507
|
Modern pretrained models can be adapted to new tasks with remarkably little data, enabling downstream applications for tasks with only tens or hundreds of examples. However, an important but neglected challenge that is especially salient in few-shot settings is task ambiguity, when the desired behavior is not uniquely specified by the provided examples. Task ambiguity can manifest in different ways: For example, the class-relevant features of an input (e.g., the shape of an object in Figure) may be spuriously correlated with other features predictive of the training labels (e.g., the color of the object), making the desired task unclear. In addition, task ambiguity may arise from an underdiverse training set, causing models to be unsure of the desired behavior for minority groups or during distribution shifts that occur at test-time.
We consider whether the task ambiguity problem can be addressed through active learning, where models select informative examples for users to label. In principle, active learning allows models themselves to assist in resolving task ambiguity by identifying examples whose labels would be informative; for example, in fig., asking for the label of the blue square helps determine that the desired task is to predict the object shape not the object color. Since it may be challenging in general for users to intuit possible sources of task ambiguity, much less how to address them with example selection, an automated active learning approach may be desirable.
In traditional settings with smaller, unpretrained models, several challenges often prevent active learning from seeing success in practice, including label noise, unlearnable examples, and a lack of generalizability across models and tasks. However, using pretrained models in few-shot or low-data settings has several key differences from traditional ML pipelines which may make them particularly well-suited for active learning. First, the effect of an individual data point (and the difference between an informative vs. an uninformative one) is magnified, compared to the thousands or millions of examples typically acquired through standard active learning pipelines. Second, pretrained models excel at learning high-level representations of inputs, which better surface relevant features and may encourage active learning to select examples that disambiguate between these high-level features (e.g. shape and color).
We consider the use of active learning on a range of spuriously correlated, imbalanced, or domain shifted datasets where task ambiguity is salient. We compare an active learning approach with a random-sampling baseline, and compare the difference in performance with and without the use of pretrained models. Our contributions are:
1. Identifying and motivating task ambiguity as a unified concern across a range of real-world datasets, and an especially salient problem for low-data settings.
2. Proposing active learning as a potential solution, along with a simple yet practical recipe that does not require adjustment for new datasets or tuning on validation data.
3. Showing that active learning can enable large gains (up to 6× reduction in data points, +12% absolute gain for the same labeling budget) and presenting scaling trends demonstrating that this is an emergent property of pretraining.
To examine the impact of task ambiguity from the perspective of the unlabeled dataset, we construct variants of the waterbirds dataset where the percentage of mismatched examples range from 95% to 50%. 4 We then proceed with AL and report results in Figure. 5 We find significantly larger gains on versions of the dataset with smaller fraction of mismatched examples (gains average 4% for 50-70% shift, and 10% for 75-95% shift) keeping. This provides stronger evidence that active learning is actually helping the model identify the true task.
However, an important but neglected challenge that is especially salient in few-shot settings is task ambiguity, when the desired behavior is not uniquely specified by the provided examples.****To examine the impact of task ambiguity from the perspective of the unlabeled dataset, we construct variants of the waterbirds dataset where the percentage of mismatched examples range from 95% to 50%.
|
Why does the random baseline make small progress even with hundreds of datapoints? + Relation to train-test mismatch.
|
This is correct! Pretraining is no cure-all, and we believe applications that feature task ambiguity (which may manifest through distribution shift) are one of the most important outstanding challenges. To capture the precise effect of these shifts, we conduct a "dose-response" experiment where we vary the fraction of mismatched examples in Waterbirds from 5% to 50%. See the main comment for more details.
|
null | false
| 64
|
Deep learning systems have shown a lot of promise for extractive Question Answering (QA), with performance comparable to humans when large scale data is available. However, practitioners looking to build QA systems for specific applications may not have the resources to collect tens of thousands of questions on corpora of their choice. At the same time, state-of-the-art machine reading systems do not lend well to low-resource QA settings where the number of labeled question-answer pairs are limited (c.f. Table 2 ). Semi-supervised QA methods like BIBREF0 aim to improve this performance by leveraging unlabeled data which is easier to collect.
In this work, we present a semi-supervised QA system which requires the end user to specify a set of base documents and only a small set of question-answer pairs over a subset of these documents. Our proposed system consists of three stages. First, we construct cloze-style questions (predicting missing spans of text) from the unlabeled corpus; next, we use the generated clozes to pre-train a powerful neural network model for extractive QA BIBREF1 , BIBREF2 ; and finally, we fine-tune the model on the small set of provided QA pairs.
Our cloze construction process builds on a typical writing phenomenon and document structure: an introduction precedes and summarizes the main body of the article. Many large corpora follow such a structure, including Wikipedia, academic papers, and news articles. We hypothesize that we can benefit from the un-annotated corpora to better answer various questions – at least ones that are lexically similar to the content in base documents and directly require factual information.
We apply the proposed system on three datasets from different domains – SQuAD BIBREF3 , TriviaQA-Web BIBREF4 and the BioASQ challenge BIBREF5 . We observe significant improvements in a low-resource setting across all three datasets. For SQuAD and TriviaQA, we attain an F1 score of more than 50% by merely using 1% of the training data. Our system outperforms the approaches for semi-supervised QA presented in BIBREF0 , and a baseline which uses the same unlabeled data but with a language modeling objective for pretraining. In the BioASQ challenge, we outperform the best performing system from previous year's challenge, improving over a baseline which does transfer learning from the SQuAD dataset. Our analysis reveals that questions which ask for factual information and match to specific parts of the context documents benefit the most from pretraining on automatically constructed clozes.
Our analysis reveals that questions which ask for factual information and match to specific parts of the context documents benefit the most from pretraining on automatically constructed clozes.
|
What does their analysis reveal?
|
Questions which ask for factual information and match to specific parts of the context documents benefit the most from pretraining on automatically constructed clozes.
|
1810.02100
| false
| null |
FLOAT SELECTED: Table 2.3: Labelled attachment scores achieved by the MST, Malt, and Mate parsers trained on the Conll training set and tested on different domains.
We further evaluate our approach on our main evaluation corpus. The method is tested on both in-domain and out-of-domain parsing. Our DLM-based approach achieved large improvement on all five domains evaluated (Conll, Weblogs, Newsgroups, Reviews, Answers). We achieved the labelled and unlabelled improvements of up to 0.91% and 0.82% on Newsgroups domain. On average we achieved 0.6% gains for both labelled and unlabelled scores on four out-of-domain test sets. We also improved the in-domain accuracy by 0.36% (LAS) and 0.4% (UAS).
FLOAT SELECTED: Table 2.3: Labelled attachment scores achieved by the MST, Malt, and Mate parsers trained on the Conll training set and tested on different domains.
We further evaluate our approach on our main evaluation corpus. The method is tested on both in-domain and out-of-domain parsing. Our DLM-based approach achieved large improvement on all five domains evaluated (Conll, Weblogs, Newsgroups, Reviews, Answers).
|
Which English domains do they evaluate on?
|
The answers are shown as follows:
* Conll, Weblogs, Newsgroups, Reviews, Answers
|
1707.00110
| false
| null |
FLOAT SELECTED: Table 1: BLEU scores and computation times with varyingK and sequence length compared to baseline models with and without attention.
FLOAT SELECTED: Table 1: BLEU scores and computation times with varyingK and sequence length compared to baseline models with and without attention.
|
How much is the BLEU score?
|
Ranges from 44.22 to 100.00 depending on K and the sequence length.
|
null | false
| null |
Grosvenor Museum is a museum in Chester, Cheshire, in the United Kingdom. It is recorded in the National Heritage List for England as a designated Grade II listed building. Its full title is The Grosvenor Museum of Natural History and Archaeology, with Schools of Science and Art, for Chester, Cheshire and North Wales. It takes its name from the family name of the Dukes of Westminster, who are major landowners in Cheshire. The museum opened in 1886, it was extended in 1894, and major refurbishments took place between 1989 and 1999. Its contents include archaeological items from the Roman period, paintings, musical instruments, and a room arranged as a Victorian parlour.
|
List all the years in order of appearance, separated by commas.
|
1886, 1894, 1989, 1999
|
null | false
| null |
What are five totally distinct, creative ways to describe the usefulness of the police?
|
1. The police serve as a "watchful guardian" to keep us safe.
2. The police are a "protector of the law" who protects us from harm.
3. The police are an "impartial enforcer" of the law.
4. The police are an "intermediary of justice" who helps us get justice.
5. The police are a "counselor of the law" who helps us understand the law.
|
|
null | false
| null |
Why do humans like dogs?
|
Humans like dogs because they have evolved together over at least 18,000 years. During this time, dogs have learned to identify and understand complex human expressions, including facial cues and vocal language and pitch. In turn, dog physiology and cognitive development has evolved the facial muscle sets and vocalizations and cognitive communication skills which enable them to communicate their states of emotional and physical being.
|
|
null | false
| null |
ByteDance planned on Douyin expanding overseas. The founder of ByteDance, Zhang Yiming, stated that "China is home to only one-fifth of Internet users globally. If we don't expand on a global scale, we are bound to lose to peers eyeing the four-fifths. So, going global is a must."
The app was launched as TikTok in the international market in September 2017. On 23 January 2018, the TikTok app ranked first among free application downloads on app stores in Thailand and other countries.
TikTok has been downloaded more than 130 million times in the United States and has reached 2 billion downloads worldwide, according to data from mobile research firm Sensor Tower (those numbers exclude Android users in China).
In the United States, celebrities, including Jimmy Fallon and Tony Hawk, began using the app in 2018. Other celebrities, including Jennifer Lopez, Jessica Alba, Will Smith, and Justin Bieber joined TikTok as well as many others.
In January 2019, TikTok allowed creators to embed merchandise sale links into their videos.
On 3 September 2019, TikTok and the U.S. National Football League (NFL) announced a multi-year partnership. The agreement occurred just two days before the NFL's 100th season kick-off at Soldier Field, where TikTok hosted activities for fans in honor of the deal. The partnership entails the launch of an official NFL TikTok account, which is to bring about new marketing opportunities such as sponsored videos and hashtag challenges. In July 2020, TikTok, excluding Douyin, reported close to 800 million monthly active users worldwide after less than four years of existence.
|
From the passage note down the various celebrities who joined TikTok . List the results in comma separated format.
|
Jimmy Fallon, Tony Hawk, Jennifer Lopez, Jessica Alba, Will Smith, Justin Bieber
|
null | false
| null |
What is brackish water and where does it occur?
|
Brackish water is water which is a mixture of freshwater and seawater, and is often found in bays or estuaries
|
|
null | false
| null |
The coxswain (or simply the cox) is the member who steers the boat using rudder strings, and coordinates the power and rhythm of the rowers, by communicating to the crew, often through a device called a cox box and speakers. The cox usually sits in the stern of the boat facing the rowers but in bowloaders, usually seen in the coxed four and coxed pair types of boat, the coxswain lies in the bow.
|
What is the role of the coxswain?
|
A coxswain sits in the stern of a rowing boat. A coxswain steers using rudder strings and coordinates the rowers through speakers located throughout the boat.
|
null | false
| null |
Give me an example of a business model with a mission statement.
|
Mission Statement:
Transform the transportation industry with improved customer experience, efficient operations, and reliable services with cost-effectiveness and minimal environmental impact.
This will be a good mission statement for Uber as it represents the transportation industry as a whole in which Uber can make an impact. Also, we wanted to be clear about the firm’s north star, which is improved customer experience, efficient and reliable operations with low cost, and minimal environmental impact.
Uber’s business model:
Value proposition:
Uber’s value prop is to offer transportation services via internet orderings from the customer. Customers order a transportation service via a mobile-enabled app and get instant or real-time updates about the order status. This significantly differs from the current taxi-cab model, where customers must hire a taxi-cab by waiting in line at the airport or taxi station. The existing taxi hire model has a poor experience for customers and comes with a high cost.
Activities, resources, and capabilities:
The critical activity for Uber in this business model is to connect the passengers and drivers on a demand basis concerning a location with a computer algorithm and deliver the service over a mobile app for both drivers and passengers.
Uber needs computational resources and engineers to build the algorithm. Also, Uber needs to develop capabilities to attract drivers and passengers to its platform for each market, country, and city.
Revenue model:
In this model, Uber customers will pay per hire, which will include the transportation charge, driver’s labor fees, and the driver’s car maintenance, including fuel and running and maintenance expenses of the car, and a small commission for Uber that will be the revenue for the company. On average per ride, Uber expects to earn a 3% commission for basic taxi-hire service. Top-of-the-line services like Uber-X may earn up to 10% commission.
Scope of the enterprise:
This business model's primary customer segment is millennials with smartphones with internet-enabled services.
This business model's horizontal scope can be vastly expanded to different products. Uber can start with a standard taxi-hire service, and for the luxury segment, it can offer Uber-X and Uber-SUV. Uber can also branch to adjacent industries like food delivery with Uber Eats and freight delivery with Uber Freight.
The business’s sheer scope is to develop the in-house demand algorithm for connecting drivers and passengers. Uber does not intend to employ drivers. Instead, it will lease/contract the drivers on a per-ride basis and pay the contract cost to the drivers. In contrast, the computational and other IT infrastructure services will be leased/rented on services like Amazon Web Services or Microsoft Azure. This will enable Uber to be agile and move fast in various markets while protecting its intellectual property, the algorithm.
|
|
null | false
| null |
LinkedIn (/lɪŋktˈɪn/) is a business and employment-focused social media platform that works through websites and mobile apps. It launched on May 5, 2003. It is now owned by Microsoft. The platform is primarily used for professional networking and career development, and allows jobseekers to post their CVs and employers to post jobs. From 2015 most of the company's revenue came from selling access to information about its members to recruiters and sales professionals. Since December 2016, it has been a wholly owned subsidiary of Microsoft. As of March 2023, LinkedIn has more than 900 million registered members from over 200 countries and territories.
LinkedIn allows members (both workers and employers) to create profiles and connect with each other in an online social network which may represent real-world professional relationships. Members can invite anyone (whether an existing member or not) to become a connection. LinkedIn can also be used to organize offline events, join groups, write articles, publish job postings, post photos and videos, and more.
|
Please summarize what Linkedin does.
|
Linkedin is a social platform that business professionals create profiles on and network with other business professionals. It is used to network, career development, and for jobseekers to find jobs. Linkedin has over 900 million users from over 200 countries. Linkedin can be used to post resumes/CVs, organizing events, joining groups, writing articles, publishing job postings, posting picture, posting videos, and more!
|
null | false
| null |
The bench press, or chest press, is a weight training exercise where the trainee presses a weight upwards while lying on a weight training bench. Although the bench press is a full-body exercise, the muscles primarily used are the pectoralis major, the anterior deltoids, and the triceps, among other stabilizing muscles. A barbell is generally used to hold the weight, but a pair of dumbbells can also be used
|
In relation to working out, what is a bench press, and how do you perform it?
|
The bench press is a weight training exercise targeting the chest and triceps. The trainee begins by lying down on a flat bench with a straight barbell directly on top of them. The trainee lifts the bar to extend directly above the head, and then slowly descends the bar until it touches their chest. The bar is then pressed upward until the arms are once again fully extended.
|
null | false
| null |
The 2017 Southeastern Conference football season was the 85th season of SEC football and took place during the 2017 NCAA Division I FBS football season. The season began on August 31 and will end with the 2017 SEC Championship Game on December 2. The SEC is a Power Five conference under the College Football Playoff format along with the Atlantic Coast Conference, the Big 12 Conference, the Big Ten Conference, and the Pac–12 Conference. For the 2017 season, the SEC has 14 teams divided into two divisions of seven each, named East and West.
|
Tell me about the 2017 Southeastern Conference football season
|
The 2017 Southeastern Conference football season was the 85th season of SEC football and took place during the 2017 NCAA Division I FBS football season. The season began on August 31 and will end with the 2017 SEC Championship Game on December 2.
|
null | false
| null |
Walton Casuals Football Club was a semi-professional football club based in Walton-on-Thames, Surrey. Founded in 1948, the club was affiliated to the Surrey County Football Association, and had been a part of the football pyramid since 1992.
The club joined the Surrey Intermediate League upon formation, and remained there until the league disbanded in 1965. A move to the Surrey Senior League lasted just two years before becoming founding members of the Suburban League. Walton Casuals won their first major honour in 1983 as champions of the Suburban League Southern Section. In 1992, the club turned semi-professional and entered the football pyramid by joining the Surrey Premier League. The Stags earned promotion to the Combined Counties Football League in 1995, and the Isthmian League 10 years later. After two years of groundsharing, the club moved into the Elmbridge Sports Hub – a £20 million sports complex development – for the 2017–18 season. In their first season at the new stadium, Walton secured a top six finish on goal difference and went on to win the play-offs.
Walton Casuals' most notable rivalry came with Combined Counties Premier Division club Walton & Hersham, due to their close proximity. The club also had a rivalry with Molesey, who play in the Isthmian League South Division. The two teams previously competed in an annual pre-season fixture for the Mick Burgess Memorial Trophy, but the competition was halted when Molesey earned promotion to the same league in 2015.
|
Tell me about the Walton Football Club
|
The Walton Casuals Football Club was a semi-professional football club based in Walton-on-Thames, Surrey. Founded in 1948, the club was affiliated to the Surrey County Football Association, and had been a part of the football pyramid since 1992.
The club joined the Surrey Intermediate League upon formation, and remained there until the league disbanded in 1965. A move to the Surrey Senior League lasted just two years before becoming founding members of the Suburban League. Walton Casuals won their first major honour in 1983 as champions of the Suburban League Southern Section. In 1992, the club turned semi-professional and entered the football pyramid by joining the Surrey Premier League. The Stags earned promotion to the Combined Counties Football League in 1995, and the Isthmian League 10 years later. After two years of groundsharing, the club moved into the Elmbridge Sports Hub – a £20 million sports complex development – for the 2017–18 season. In their first season at the new stadium, Walton secured a top six finish on goal difference and went on to win the play-offs.
Walton Casuals' most notable rivalry came with Combined Counties Premier Division club Walton & Hersham, due to their close proximity. The club also had a rivalry with Molesey, who play in the Isthmian League South Division. The two teams previously competed in an annual pre-season fixture for the Mick Burgess Memorial Trophy, but the competition was halted when Molesey earned promotion to the same league in 2015.
|
null | false
| null |
What are some interesting things to do in the spa town of Harrogate, North Yorkshire?
|
Some interesting things to do in the spa town of Harrogate, North Yorkshire, are to visit the Royal Pump Room Museum, take a stroll through the Valley Gardens, and have afternoon tea at Bettys Cafe Tea Rooms.
|
|
null | false
| null |
What are some things to do in Rochester, NY?
|
- Eat a Garbage Plate, a local specialty
- Visit the Eastman House, the world's oldest museum dedicated to photography
- Spend a day at the Museum of Play, a museum dedicated to exploration and play
- See the Lamberton Conservatory, a botanical garden with walking trails
- Watch a Red Wings game, Rochester's minor league baseball team
|
|
null | false
| null |
Classify each of these bands as Polish death metal, Norwegian death metal, or K-Pop: BTS, EXO, Red Velvet, Cadaver, Darkthrone, Behemoth, Vader, Empatic, Girls' Generation, TWICE, Blood Red Throne, Emperor, Gorgoroth, Decapitated
|
Polish death metal bands: Behemoth, Vader, Empatic, Decapitated
Norwegian death metal bands: Cadaver, Darkthrone, Blood Red Throne, Emperor, Gorgoroth
K-Pop bands: BTS, EXO, Red Velvet, Girls' Generation, TWICE
|
|
null | false
| null |
Tell me whether these are TV shows or books: Modern Family, You, Don Quixote, A Tale of Two Cities, The Little Prince
|
Modern Family (TV Show), You (TV Show), Don Quixote (Book), A Tale of Two Cities (Book), The Little Prince (Book)
|
|
null | false
| null |
Tell me whether these are movies or books: Gone with the wind, The Canterville Ghost, Battleship
|
book and movie, book, movie
|
|
null | false
| null |
The addition of pineapple to the traditional mix of tomato sauce and cheese, along with either ham or bacon, later became popular locally and eventually became a staple offering of pizzerias on a global scale. The name of this creation is, in fact, actually not directly inspired by the U.S. state of Hawaii at all; Panopoulos chose the name Hawaiian after the brand of canned pineapple they were using at the time.
|
Why is it called the Hawaiian pizza?
|
The name comes from the canned pineapple brand
|
null | false
| null |
Which one of these are unlike the others: Toyota Supra, Mitsubishi 3000GT, Mazda RX-7, Nissan 300ZX?
|
All of these are iconic 90s Japanese sports cars, but the RX-7 is unique in that it has a rotary engine, as opposed to the piston engines of the other cars.
|
|
null | false
| null |
How should I go about meeting a romantic life partner?
|
There is no single way to go about meeting a romantic life partner however, here are some ideas. You can sign up for an online dating application and try your luck at swiping on potential matches. If that doesn't suit your fancy, you can try visiting cafes, bookstores and stores in hopes of randomly running into your person. Other popular methods include: being introduced by friends or family, trying a group activity that you find fun and engaging or simply letting luck/happenstance intervene.
|
|
null | false
| null |
Nobo is a manufacturing company in Norway. It started as NOrsk BOttefabrikk (the Norwegian Bucket Factory) in Lade, Trondheim in 1918 - the name Nobo is a contraction of the original title. The factory originally produced only buckets, but later made several sheet iron products as well. It now manufactures desks, electric wall heaters and filing cabinets. The company developed and a new factory was built in Stjordal. Nobo then produced electric heating accessories only, which it is still doing today.
In the 1990s the original Nobø factory was closed down and the area is now a housing estate of about 220 houses and flats, called Ingemann Torps Street or Nobø-tomta (Nobø-field). The area was planned by Heimdal Utbyggingsselskap and Pir II Architects.
|
Give me a summary about the company Nobo
|
Nobo is a manufacturing company in Norway. It started as NOrsk BOttefabrikk (the Norwegian Bucket Factory) in Lade, Trondheim in 1918 - the name Nobo is a contraction of the original title. The factory originally produced only buckets, but later made several sheet iron products as well. It now manufactures desks, electric wall heaters and filing cabinets. The company developed and a new factory was built in Stjordal. Nobo then produced electric heating accessories only, which it is still doing today.
In the 1990s the original Nobo factory was closed down and the area is now a housing estate of about 220 houses and flats, called Ingemann Torps Street or Nobo-tomta (Nobo-field). The area was planned by Heimdal Utbyggingsselskap and Pir II Architects.
|
null | false
| null |
Write a letter to my mum and dad detailing the reasons why they should buy me a pet tortoise
|
Dear Mum and Dad,
As you know, I have wanted a pet tortoise for some time now. I have given it a lot of thought and done a great deal of research into what might be involved. I am aware that having a pet is a huge commitment so I am writing to you to explain why I think you should buy me a pet tortoise and why a tortoise is a fantastic choice of pet.
Looking after a pet tortoise will give me a strong sense of responsibility. I know that I will have to ensure my tortoise is well fed, kept at the right temperature, given opportunities to bask under a heated lamp or in the sunshine if warm enough. I am very willing to give the time and commitment necessary to look after my tortoise. The tortoise will be my sole responsibility and I feel mature enough to commit to its care and happiness.
As you know, I suffer from anxiety and school makes me particularly worried sometimes. It has been shown through numerous studies that pets reduce anxiety and stress in humans. I feel sure that having a tortoise will be a fantastic distraction from my worries and bring me great enjoyment. Contrary to popular belief, tortoises love having their shells stroked and their chins tickled and I think it would be very relaxing for me to pet a tortoise when I get home from school.
Tortoises are very self-sufficient. Provided they are given a warm place to bask, access to fresh water and plenty of fresh food such as dandelions, kale, lettuce and other leafy vegetables, they are very happy to be left to their own devices. As we live in the UK where temperatures can be cool, we would need to purchase a tortoise table with good substrate, a basking lamp and some hides where the tortoise can tuck itself away. We have lots of space for this in our kitchen. When it is warm, the tortoise can go into our garden in an enclosure. A tortoise table and lamp will cost about £200 but this is just an initial outlay. Thereafter, keeping a tortoise is very cheap. In fact, you can pick most of their food for free by foraging which I am happy to do when we go for families walk.
Contrary to popular belief, tortoises can be very friendly. As I have mentioned above, they love being stroked but did you also know that they can recognise their owners and some can even respond to their names? In fact, a study showed that tortoises can recognise their previous owners that they haven’t seen in years so they are both friendly and very intelligent.
You might be worried about hibernating a tortoise but I have done lots of research on this and now know exactly what to do. If we buy a tortoise, it will need to hibernate for a maximum of three months from the end of September. We will need to hibernate our tortoise in a well-insulated box. The temperature mustn’t go above or below 5 degrees celsius and our garage would be the perfect place for it. We have a temperature gauge in there so I can keep an eye on the heat levels.
Finally, I have looked into the cost of buying a tortoise. I would suggest we buy a tortoise which is a couple of years’ old as they require less care. We can buy a preloved tortoise which has been microchipped for about £100. And when we go on holiday, I have found a Tortoise Hotel which is run by a vet and costs just £60 for a 10 day stay.
I think you will agree that I have conducted very thorough research into looking after a tortoise and I hope you will agree that I am now mature enough to look after a pet. I truly believe that a pet tortoise will give me a strong sense of responsibility and will have a positive calming effect on my mood. Indeed, I think the whole family will enjoy having a tortoise. They are friendly, loving, interesting and intelligent animals and we will all benefit from having one.
I would be grateful if you would give my request full consideration.
Lots of Love,
Wilf x
|
|
null | false
| null |
Which characters belong to DC or Marvel Universe? Susan Storm, Green Lantern
|
Susan Storm is Marvel, Green Lantern is DC
|
|
1911.12569
| false
| null |
FLOAT SELECTED: TABLE III COMPARISON WITH THE STATE-OF-THE-ART SYSTEMS OF SEMEVAL 2016 TASK 6 ON SENTIMENT DATASET.
FLOAT SELECTED: TABLE IV COMPARISON WITH THE STATE-OF-THE-ART SYSTEMS PROPOSED BY [16] ON EMOTION DATASET. THE METRICS P, R AND F STAND FOR PRECISION, RECALL AND F1-SCORE.
Table TABREF19 shows the comparison of our proposed system with the existing state-of-the-art system of SemEval 2016 Task 6 for the sentiment dataset. BIBREF7 used feature-based SVM, BIBREF39 used keyword rules, LitisMind relied on hashtag rules on external data, BIBREF38 utilized a combination of sentiment classifiers and rules, whereas BIBREF37 used a maximum entropy classifier with domain-specific features. Our system comfortably surpasses the existing best system at SemEval. Our system manages to improve the existing best system of SemEval 2016 task 6 by 3.2 F-score points for sentiment analysis.
We also compare our system with the state-of-the-art systems proposed by BIBREF15 on the emotion dataset. The comparison is demonstrated in Table TABREF22. Maximum entropy, SVM, LSTM, Bi-LSTM, and CNN were the five individual systems used by BIBREF15. Overall, our proposed system achieves an improvement of 5 F-Score points over the existing state-of-the-art system for emotion analysis. Individually, the proposed system improves the existing F-scores for all the emotions except surprise. The findings of BIBREF15 also support this behavior (i.e. worst result for the surprise class). This could be attributed to the data scarcity and a very low agreement between the annotators for the emotion surprise.
FLOAT SELECTED: TABLE III COMPARISON WITH THE STATE-OF-THE-ART SYSTEMS OF SEMEVAL 2016 TASK 6 ON SENTIMENT DATASET.
FLOAT SELECTED: TABLE IV COMPARISON WITH THE STATE-OF-THE-ART SYSTEMS PROPOSED BY [16] ON EMOTION DATASET. THE METRICS P, R AND F STAND FOR PRECISION, RECALL AND F1-SCORE.
Table TABREF19 shows the comparison of our proposed system with the existing state-of-the-art system of SemEval 2016 Task 6 for the sentiment dataset.
We also compare our system with the state-of-the-art systems proposed by BIBREF15 on the emotion dataset. The comparison is demonstrated in Table TABREF22
|
What are the state-of-the-art systems?
|
For sentiment analysis UWB, INF-UFRGS-OPINION-MINING, LitisMind, pkudblab and SVM + n-grams + sentiment and for emotion analysis MaxEnt, SVM, LSTM, BiLSTM and CNN
|
null | false
| null |
Mother Teresa said, "By blood, I am Albanian. By citizenship, an Indian. By faith, I am a Catholic nun. As to my calling, I belong to the world. As to my heart, I belong entirely to the Heart of Jesus." Fluent in five languages – Bengali, Albanian, Serbian, English and Hindi – she made occasional trips outside India for humanitarian reasons. At the height of the Siege of Beirut in 1982, Mother Teresa rescued 37 children trapped in a front-line hospital by brokering a temporary cease-fire between the Israeli army and Palestinian guerrillas. Accompanied by Red Cross workers, she travelled through the war zone to the hospital to evacuate the young patients. When Eastern Europe experienced increased openness in the late 1980s, Mother Teresa expanded her efforts to Communist countries which had rejected the Missionaries of Charity. She began dozens of projects, undeterred by criticism of her stands against abortion and divorce: "No matter who says what, you should accept it with a smile and do your own work." She visited Armenia after the 1988 earthquake and met with Soviet Premier Nikolai Ryzhkov. Mother Teresa travelled to assist the hungry in Ethiopia, radiation victims at Chernobyl and earthquake victims in Armenia. In 1991 she returned to Albania for the first time, opening a Missionaries of Charity Brothers home in Tirana. By 1996, the Missionaries of Charity operated 517 missions in over 100 countries. The number of sisters in the Missionaries of Charity grew from twelve to thousands, serving the "poorest of the poor" in 450 centres worldwide. The first Missionaries of Charity home in the United States was established in the South Bronx area of New York City, and by 1984 the congregation operated 19 establishments throughout the country.
|
Given the reference text about Mother Teresa's international charity, how many languages did Mother Teresa speak?
|
Mother Teresa was fluent in five languages, namely Bengali, Albanian, Serbian, English and Hindi.
|
1908.04917
| false
| null |
As shown in Equation ( EQREF6 ), tone prediction sub-network ( INLINEFORM0 ) takes video and pinyin sequence as inputs and predict corresponding tone sequence. This problem is modeled as a sequence-to-sequence learning problem too. The corresponding model architecture is shown in Figure FIGREF8 .
In order to take both video and pinyin information into consideration when producing tone, a dual attention mechanism BIBREF3 is employed. Two independent attention mechanisms are used for video and pinyin sequence. Video context vectors INLINEFORM0 and pinyin context vectors INLINEFORM1 are fused when predicting a tone character at each decoder step.
The video encoder is the same as in Section SECREF7 and the pinyin encoder is: DISPLAYFORM0
The pinyin prediction sub-network transforms video sequence into pinyin sequence, which corresponds to INLINEFORM0 in Equation ( EQREF6 ). This sub-network is based on the sequence-to-sequence architecture with attention mechanism BIBREF8 . We name the encoder and decoder the video encoder and pinyin decoder, for the encoder process video sequence, and the decoder predicts pinyin sequence. The input video sequence is first fed into the VGG model BIBREF9 to extract visual feature. The output of conv5 of VGG is appended with global average pooling BIBREF10 to get the 512-dim feature vector. Then the 512-dim feature vector is fed into video encoder. The video encoder can be denoted as: DISPLAYFORM0
As shown in Equation ( EQREF6 ), tone prediction sub-network ( INLINEFORM0 ) takes video and pinyin sequence as inputs and predict corresponding tone sequence.
Video context vectors INLINEFORM0 and pinyin context vectors INLINEFORM1 are fused when predicting a tone character at each decoder step.
The video encoder is the same as in Section SECREF7 and the pinyin encoder is: DISPLAYFORM0
The input video sequence is first fed into the VGG model BIBREF9 to extract visual feature. The output of conv5 of VGG is appended with global average pooling BIBREF10 to get the 512-dim feature vector. Then the 512-dim feature vector is fed into video encoder.
|
What visual information characterizes tones?
|
The answers are shown as follows:
* video sequence is first fed into the VGG model BIBREF9 to extract visual feature
|
null | false
| null |
Stefano "The Undertaker" Magaddino (Italian pronunciation: [ˈsteːfano maɡadˈdiːno]; October 10, 1891 – July 19, 1974) was an Italian-born crime boss of the Buffalo crime family in western New York. His underworld influence stretched from Ohio to Southern Ontario and as far east as Montreal, Quebec. Known as Don Stefano to his friends and The Undertaker to others, he was also a charter member of the American Mafia's ruling council, The Commission.
|
Please list the nicknames of the famed crime boss Stefano Magaddino, separated by a comma.
|
Don Stefano, The Undertaker
|
null | false
| 2
|
There has been significant progress on Named Entity Recognition (NER) in recent years using models based on machine learning algorithms BIBREF0 , BIBREF1 , BIBREF2 . As with other Natural Language Processing (NLP) tasks, building NER systems typically requires a massive amount of labeled training data which are annotated by experts. In real applications, we often need to consider new types of entities in new domains where we do not have existing annotated data. For such new types of entities, however, it is very hard to find experts to annotate the data within short time limits and hiring experts is costly and non-scalable, both in terms of time and money.
In order to quickly obtain new training data, we can use crowdsourcing as one alternative way at lower cost in a short time. But as an exchange, crowd annotations from non-experts may be of lower quality than those from experts. It is one biggest challenge to build a powerful NER system on such a low quality annotated data. Although we can obtain high quality annotations for each input sentence by majority voting, it can be a waste of human labors to achieve such a goal, especially for some ambiguous sentences which may require a number of annotations to reach an agreement. Thus majority work directly build models on crowd annotations, trying to model the differences among annotators, for example, some of the annotators may be more trustful BIBREF3 , BIBREF4 .
Here we focus mainly on the Chinese NER, which is more difficult than NER for other languages such as English for the lack of morphological variations such as capitalization and in particular the uncertainty in word segmentation. The Chinese NE taggers trained on news domain often perform poor in other domains. Although we can alleviate the problem by using character-level tagging to resolve the problem of poor word segmentation performances BIBREF5 , still there exists a large gap when the target domain changes, especially for the texts of social media. Thus, in order to get a good tagger for new domains and also for the conditions of new entity types, we require large amounts of labeled data. Therefore, crowdsourcing is a reasonable solution for these situations.
In this paper, we propose an approach to training a Chinese NER system on the crowd-annotated data. Our goal is to extract additional annotator independent features by adversarial training, alleviating the annotation noises of non-experts. The idea of adversarial training in neural networks has been used successfully in several NLP tasks, such as cross-lingual POS tagging BIBREF6 and cross-domain POS tagging BIBREF7 . They use it to reduce the negative influences of the input divergences among different domains or languages, while we use adversarial training to reduce the negative influences brought by different crowd annotators. To our best knowledge, we are the first to apply adversarial training for crowd annotation learning.
In the learning framework, we perform adversarial training between the basic NER and an additional worker discriminator. We have a common Bi-LSTM for representing annotator-generic information and a private Bi-LSTM for representing annotator-specific information. We build another label Bi-LSTM by the crowd-annotated NE label sequence which reflects the mind of the crowd annotators who learn entity definitions by reading the annotation guidebook. The common and private Bi-LSTMs are used for NER, while the common and label Bi-LSTMs are used as inputs for the worker discriminator. The parameters of the common Bi-LSTM are learned by adversarial training, maximizing the worker discriminator loss and meanwhile minimizing the NER loss. Thus the resulting features of the common Bi-LSTM are worker invariant and NER sensitive.
For evaluation, we create two Chinese NER datasets in two domains: dialog and e-commerce. We require the crowd annotators to label the types of entities, including person, song, brand, product, and so on. Identifying these entities is useful for chatbot and e-commerce platforms BIBREF8 . Then we conduct experiments on the newly created datasets to verify the effectiveness of the proposed adversarial neural network model. The results show that our system outperforms very strong baseline systems. In summary, we make the following contributions:
In the learning framework, we perform adversarial training between the basic NER and an additional worker discriminator.
|
How did the authors perform adversarial training?
|
The authors perform adversarial training between the basic NER and an additional worker discriminator.
|
2004.01894
| false
| null |
In the unsupervised scenario similarity is computed as the cosine of the produced $h_{L}$ and $h_{R}$ sentence/image representations.
In the unsupervised scenario similarity is computed as the cosine of the produced $h_{L}$ and $h_{R}$ sentence/image representations.
|
How they compute similarity between the representations?
|
The answers are shown as follows:
* similarity is computed as the cosine of the produced $h_{L}$ and $h_{R}$ sentence/image representations
|
null | false
| null |
What are some benefits of running?
|
Running has many benefits for body and mind. Some benefits are:
1. Improves cardiovascular health: running strengthens your heard by enabling it to pump more blood volume per beat, and your lungs become more capable of taking in more air per breath.
2. Increases bone density: high impact exercises, like running, place stress on bones that strengthens their structure
3. Reduces stress: running can reduce the stress hormone, cortisol
4. Improves your mood: running helps boost your mood, concentration, and overall quality of life
5. Improves sleep: cardio exercises, such as running, can help you fall asleep faster and improve overall sleep quality
6. Burns calories: running is metabolically demanding and burns a lot of calories
|
|
null | false
| null |
Camp Nou (Catalan pronunciation: [ˌkamˈnɔw], meaning new field, often referred to in English as the Nou Camp), officially branded as Spotify Camp Nou for sponsorship reasons, is a football stadium in Barcelona, Spain. It has been the home stadium of FC Barcelona since its completion in 1957. With a current seating capacity of 99,354, it is the largest stadium in Spain and Europe, and the second largest association football stadium in the world.
It has hosted two European Cup/Champions League finals in 1989 and 1999, two European Cup Winners' Cup finals, four Inter-Cities Fairs Cup final games, five UEFA Super Cup games, four Copa del Rey finals, two Copa de la Liga finals, and twenty-one Supercopa de España finals. It also hosted five matches in the 1982 FIFA World Cup (including the opening game), two out of four matches at the 1964 European Nations' Cup, and the football tournament's final at the 1992 Summer Olympics.
On 15 March 2022, it was announced that music streaming service Spotify had reached a deal with FC Barcelona to acquire the naming rights to the stadium in a deal worth $310 million. Following the approval of the sponsorship agreement with Spotify by FC Barcelona's Extraordinary Assembly of Delegate Members on 3 April 2022, the stadium was officially renamed on 1 July 2022 as Spotify Camp Nou.
In April 2022, it was announced that renovation of the stadium will commence in June 2022 after the season's end.
|
Which is the largest football stadium in Europe
|
The Camp Nou, meaning new field, often referred to in English as the Nou Camp), officially branded as Spotify Camp Nou for sponsorship reasons, is a football stadium in Barcelona, Spain. It has been the home stadium of FC Barcelona since its completion in 1957. With a current seating capacity of 99,354, it is the largest stadium in Spain and Europe, and the second largest association football stadium in the world.
It has hosted two European Cup/Champions League finals in 1989 and 1999, two European Cup Winners' Cup finals, four Inter-Cities Fairs Cup final games, five UEFA Super Cup games, four Copa del Rey finals, two Copa de la Liga finals, and twenty-one Supercopa de España finals. It also hosted five matches in the 1982 FIFA World Cup (including the opening game), two out of four matches at the 1964 European Nations' Cup, and the football tournament's final at the 1992 Summer Olympics.
On 15 March 2022, it was announced that music streaming service Spotify had reached a deal with FC Barcelona to acquire the naming rights to the stadium in a deal worth $310 million. Following the approval of the sponsorship agreement with Spotify by FC Barcelona's Extraordinary Assembly of Delegate Members on 3 April 2022, the stadium was officially renamed on 1 July 2022 as Spotify Camp Nou.
In April 2022, it was announced that renovation of the stadium will commence in June 2022 after the season's end.
|
null | false
| null |
Like the Battles of Marathon and Thermopylae, Salamis has gained something of a 'legendary' status (unlike, for instance, the more decisive Battle of Plataea), perhaps because of the desperate circumstances and the unlikely odds. A significant number of historians have stated that Salamis is one of the most significant battles in human history (though the same is often stated of Marathon). In a more extreme form of this argument, some historians argue that if the Greeks had lost at Salamis, the ensuing conquest of Greece by the Persians would have effectively stifled the growth of Western Civilization as we know it. This view is based on the premise that much of modern Western society, such as philosophy, science, personal freedom and democracy are rooted in the legacy of Ancient Greece. Thus, this school of thought argues that, given the domination of much of modern history by Western Civilization, Persian domination of Greece might have changed the whole trajectory of human history. It is also worth mentioning that the celebrated blossoming of hugely influential Athenian culture occurred only after the Persian wars were won.
From: https://en.wikipedia.org/wiki/Battle_of_Salamis
|
What about the battle of Salamis (fought in 480 BC), do many historians believe make it one of the most significant battles of human history?
|
The battle (fought between the Persian Empire and Greek City-States in 480 BC) is still considered by many historians to be a foundational event in the development of Western Civilization and modern Democracy. Some say that if the Greeks had lost at Salamis it could have had profound effects on the trajectory of human history. Historian's trace back much of today's Western ideals in philosophy, culture, politics, and science to Ancient Greece. They deduce that if Persia had won the Greco-Persian wars it could have oppressed the blossoming of many modern Western schools of thought.
Salamis is one of many skirmishes fought over the course of almost half a century during this conflict. There were other pivotal battles that contributed to the Greek victory, like Marathon and Plataea. However, if Salamis had been won by the Persians, the Greek navy would have been unable to stop them from establishing beachheads across Greece. Much of the Greek victory has been attributed to their use of the environment to their advantage; leveraging natural chokepoints (like at Salamis) to reduce the effectiveness of Persia's numerical advantage. Had Salamis fallen to the Persians it's reasonable to assume that this strategy would have failed and Persia's massive armies would have been able to dominate the Greeks.
It's also impossible to predict the outcome of the war if Persia had won this battle, even though chances are things would have ended very differently. Salamis is perhaps more highly regarded because of the long odds the Greek navy faced. They were able to prevail despite being largely outnumbered, which could also be why it's considered such a remarkable event.
|
null | false
| null |
What are some silly things to drink out of instead of a cup?
|
Things you could drink from that aren't cups could include a hummingbird feeder, a syrup container, a (plastic lined) mini piñata, an erlenmeyer flask, a water gun, an empty medicine bottle, a plant vase or pot, or a watering can.
|
|
null | false
| null |
What is accounting?
|
Accounting is a field of study that deals with financial data and transactions. It is a way business keep tracking of payments made to vendors, payments made to employees, taxes paid, payments made to shareholders, and payments for general expenses. At the heart of it an accountant is responsible for the accounting books for every single transaction a company partakes in - receiving and paying money.
|
|
null | false
| null |
The Old Thatch Tavern became the brewery for Stratford-upon-Avon in 1470 when it was also a pub. It has been a licensed pub since 1623 and it is claimed that it is the oldest pub in Stratford.
|
Extract all of the dates mentioned in this paragraph and list them using bullets in the format {Date} - {Description}
|
• 1470 - when the Old Thatch Tavern became a brewery
• 1623 - when the Old Thatch Tavern became a licensed pub
|
null | false
| null |
The 2005 Lexus Cup was the inaugural edition of the annual golf match competed by women representing Asia and an international squad. Each team was made up of twelve members. The competition took place at the Tanah Merah Country Club in Singapore from 9–11 December 2005. Lexus was he title sponsor while Rolex, DBS, Singapore Airlines, and Singapore Sports Council are main sponsors. The total purse was US$960,000, with $50,000 going to each member of the winning team and $30,000 to members of the other team.
|
In the 2005 Lexus Cup, how many members did each team have?
|
Each team was made up of twelve members.
|
null | false
| 23
|
Since its rise in 2013, the Islamic State of Iraq and Syria (ISIS) has utilized the Internet to spread its ideology, radicalize individuals, and recruit them to their cause. In comparison to other Islamic extremist groups, ISIS' use of technology was more sophisticated, voluminous, and targeted. For example, during ISIS' advance toward Mosul, ISIS related accounts tweeted some 40,000 tweets in one day BIBREF0.However, this heavy engagement forced social media platforms to institute policies to prevent unchecked dissemination of terrorist propaganda to their users, forcing ISIS to adapt to other means to reach their target audience.
One such approach was the publication of online magazines in different languages including English. Although discontinued now, these online resources provided a window into ISIS ideology, recruitment, and how they wanted the world to perceive them. For example, after predominantly recruiting men, ISIS began to also include articles in their magazines that specifically addressed women. ISIS encouraged women to join the group by either traveling to the caliphate or by carrying out domestic attacks on behalf of ISIS in their respective countries. This tactical change concerned both practitioners and researchers in the counterterrorism community. New advancements in data science can shed light on exactly how the targeting of women in extremist propaganda works and whether it differs significantly from mainstream religious rhetoric.
We utilize natural language processing methods to answer three questions:
What are the main topics in women-related articles in ISIS' online magazines?
What similarities and/or differences do these topics have with non-violent, non-Islamic religious material addressed specifically to women?
What kind of emotions do these articles evoke in their readers and are there similarities in the emotions evoked from both ISIS and non-violent religious materials?
As these questions suggest, to understand what, if anything, makes extremist appeals distinctive, we need a point of comparison in terms of the outreach efforts to women from a mainstream, non-violent religious group. For this purpose, we rely on an online Catholic women's forum. Comparison between Catholic material and the content of ISIS' online magazines allows for novel insight into the distinctiveness of extremist rhetoric when targeted towards the female population. To accomplish this task, we employ topic modeling and an unsupervised emotion detection method.
The rest of the paper is organized as follows: in Section SECREF2, we review related works on ISIS propaganda and applications of natural language methods. Section SECREF3 describes data collection and pre-processing. Section SECREF4 describes in detail the approach. Section SECREF5 reports the results, and finally, Section SECREF6 presents the conclusion.
To accomplish this task, we employ topic modeling and an unsupervised emotion detection method.
|
Does the author use topic modeling？
|
Yes.
|
null | false
| null |
Thysanozoon nigropapillosum is a species of polyclad flatworms belonging to the family Pseudocerotidae. Some common names include gold-speckled flatworm, marine flatworm, yellow papillae flatworm, yellow-spotted flatworm, and yellow-spotted polyclad flatworm.Thysanozoon nigropapillosum has a long body and broad shape. They grow up to 3 in (76 mm). The dorsal surface is deep black and covered with numerous yellow-tipped papillae varying in size. The ventral surface is dark brown. The outer margin of the body is slightly wavy and bordered in opaque white. They have small, ear-like pseudotentacles in the middle of the anterior end. They swim by propelling themselves through the water with a rhythmic undulating motion of the body.This species is widespread in the tropical Indo-Pacific.Thysanozoon nigropapillosum is quite common along the external reef in the shallow sub-tidal zone. It can swim by undulating and rhythmically contracting the body margins. It feeds on tunicates, using its mouth and large pharynx to engulf Didemnum spp., and later regurgitates food pellets containing the calcareous spicules present in their tunics. Polyclad flatworms are hermaphrodites, with each one of a pair of flatworms trying to inseminate the other. Transfer of sperm may be by hypodermic insemination, but this is made difficult in this species by the presence of the papillae, and dermal impregnation often occurs. This involves sperm entering the body directly after being placed in the marginal region of the skin. Such dermal insemination is facilitated by the fact that the flatworm has two penises which are strengthened by the presence of rhabdites, and these can grip the margin of the partner and keep hold, even while the recipient is swimming.World Database of Marine SpeciesSeaslugsWormsZipcodezooGwannonPhotos of Thysanozoon nigropapillosum on Sealife Collection
|
Please summarize what Thysanozoon nigropapillosum are
|
Thysanozoon nigropapillosum is a species of polyclad flatworms belonging to the family Pseudocerotidae. Thysanozoon nigropapillosum are commonly named as gold-speckled flatorm, marine flatworm, yellow paipllae flatworm, yellow spotted flatworm, and yellow spotty polyclad. They have a long body and broad shape, and can grow up to 3 inches (76mm). They are widespread in the Indo-Pacific, and feed on tunicates, and later regurgitates food pellets that contain calcareous spicules that are present in their tunics. The Polyclad flatworms are hermaphrodites, with each pair of flatworms attempt to inseminate each other.
|
null | false
| null |
Describe what it's like in Breckenridge, Colorado
|
Breckenridge, Colorado is largely known for its ski resort of the same name. Breckenridge is one of the highest-altitude tourist destinations in the United States, at an elevation of approximately 9600 feet (or 2926 meters). It has a small year-round resident population (a few thousand people), but has millions of visitors every year. Breckenridge Ski Resort operates across five peaks, and includes a chairlift that operates at 12,840 feet (the highest in North America). Although Breckenridge gets more than 300 inches of snow on average each year, it also has abundant sunshine with about 300 days being sunny. Breckenridge has cold winters and moderate summers.
|
|
null | false
| null |
List some common symptoms that someone may experience if they acquired a gastrointestinal bug.
|
Common symptoms someone who has a gastrointestinal infection may experience include cramping, diarrhea, low-grade fever, vomiting, and nausea.
|
|
2002.05104
| false
| null |
More specifically, we observe the impact of: (i) pre-trained word embeddings BIBREF11, BIBREF12, recurrent BIBREF13 and transformer-based sentence encoders BIBREF14 as question representation strategies; (ii) distinct convolutional neural networks used for visual feature extraction BIBREF15, BIBREF16, BIBREF17; and (iii) standard fusion strategies, as well as the importance of two main attention mechanisms BIBREF18, BIBREF19. We notice that even using a relatively simple baseline architecture, our best models are competitive to the (maybe overly-complex) state-of-the-art models BIBREF20, BIBREF21. Given the experimental nature of this work, we have trained over 130 neural network models, accounting for more than 600 GPU processing hours. We expect our findings to be useful as guidelines for training novel VQA models, and that they serve as a basis for the development of future architectures that seek to maximize predictive performance.
More specifically, we observe the impact of: (i) pre-trained word embeddings BIBREF11, BIBREF12, recurrent BIBREF13 and transformer-based sentence encoders BIBREF14 as question representation strategies; (ii) distinct convolutional neural networks used for visual feature extraction BIBREF15, BIBREF16, BIBREF17; and (iii) standard fusion strategies, as well as the importance of two main attention mechanisms BIBREF18, BIBREF19.
|
What type of experiments are performed?
|
The answers are shown as follows:
* pre-trained word embeddings BIBREF11, BIBREF12
* recurrent BIBREF13
* transformer-based sentence encoders BIBREF14
* distinct convolutional neural networks
* standard fusion strategies
* two main attention mechanisms BIBREF18, BIBREF19
|
null | false
| null |
James grew up playing basketball for St. Vincent–St. Mary High School in his hometown of Akron, Ohio. He was heavily touted by the national media as a future NBA superstar. A prep-to-pro, he was selected by the Cleveland Cavaliers with the first overall pick of the 2003 NBA draft. Named the 2004 NBA Rookie of the Year, he soon established himself as one of the league's premier players, leading the Cavaliers to their first NBA Finals appearance in 2007 and winning the NBA MVP award in 2009 and 2010. After failing to win a championship with Cleveland, James left in 2010 as a free agent to join the Miami Heat; this was announced in a nationally televised special titled The Decision and is among the most controversial free agency moves in sports history.
|
Given this paragraph on LeBron James' early career, when did he win his first NBA MVP?
|
LeBron James won his first NBA MVP in 2009.
|
null | false
| null |
Professor Tatyana Grigorievna Nazarenko is one of the leading Russian contemporary painters who has set new directions in Russian Art and Painting starting from the 1970s. She was born in Moscow where she still lives and works.
|
Does Grigorievna Nazarenko live in Moscow?
|
Yes, she lives in Moscow
|
null | false
| null |
A tepui /ˈtɛpwi/, or tepuy (Spanish: [teˈpuj]), is a table-top mountain or mesa found in South America, especially in Venezuela and western Guyana. The word tepui means "house of the gods" in the native tongue of the Pemon, the indigenous people who inhabit the Gran Sabana.
|
What is a Tepui?
|
A tepui /ˈtɛpwi/, or tepuy (Spanish: [teˈpuj]), is a table-top mountain or mesa found in South America, especially in Venezuela and western Guyana. The word tepui means "house of the gods" in the native tongue of the Pemon, the indigenous people who inhabit the Gran Sabana.
Tepuis tend to be found as isolated entities rather than in connected ranges, which makes them the host of a unique array of endemic plant and animal species. Some of the most outstanding tepuis are Auyantepui, Autana, Neblina, and Mount Roraima. They are typically composed of sheer blocks of Precambrian quartz arenite sandstone that rise abruptly from the jungle, giving rise to spectacular natural scenery. Auyantepui is the source of Angel Falls, the world's tallest waterfall.
|
null | false
| 107
|
We re-implemented five keyphrase extraction models : the first two are commonly used as baselines, the third is a resource-lean unsupervised graph-based ranking approach, and the last two were among the top performing systems in the SemEval-2010 keyphrase extraction task BIBREF0 . We note that two of the systems are supervised and rely on the training set to build their classification models. Document frequency counts are also computed on the training set. Stemming is applied to allow more robust matching. The different keyphrase extraction models are briefly described below:
Each model uses a distinct keyphrase candidate selection method that provides a trade-off between the highest attainable recall and the size of set of candidates. Table summarizes these numbers for each model. Syntax-based selection heuristics, as used by TopicRank and WINGNUS, are better suited to prune candidates that are unlikely to be keyphrases. As for KP-miner, removing infrequent candidates may seem rather blunt, but it turns out to be a simple yet effective pruning method when dealing with long documents. For details on how candidate selection methods affect keyphrase extraction, please refer to BIBREF16 .
Apart from TopicRank that groups similar candidates into topics, the other models do not have any redundancy control mechanism. Yet, recent work has shown that up to 12% of the overall error made by state-of-the-art keyphrase extraction systems were due to redundancy BIBREF6 , BIBREF17 . Therefore as a post-ranking step, we remove redundant keyphrases from the ranked lists generated by all models. A keyphrase is considered redundant if it is included in another keyphrase that is ranked higher in the list.
Each model uses a distinct keyphrase candidate selection method that provides a trade-off between the highest attainable recall and the size of set of candidates.
|
What about the keyphrase candidate selection method of the models?
|
Each model uses a distinct keyphrase candidate selection method that provides a trade-off between the highest attainable recall and the size of set of candidates.
|
null | false
| null |
The 1968 United States elections were held on November 5, and elected members of the 91st United States Congress. The election took place during the Vietnam War, in the same year as the Tet Offensive, the assassination of Martin Luther King, Jr., the assassination of Robert F. Kennedy, and the protests of 1968. The Republican Party won control of the presidency, and picked up seats in the House and Senate, although the Democratic Party retained control of Congress.
In the presidential election, Republican former Vice President Richard Nixon defeated Democratic incumbent Vice President Hubert Humphrey. Nixon won the popular vote by less than one point, but took most states outside the Northeast, and comfortably won the electoral vote. Former Alabama Governor George Wallace, of the American Independent Party, took 13.5% of the popular vote, and won the electoral votes of the Deep South. After incumbent Democratic President Lyndon B. Johnson declined to seek re-election, Humphrey won the Democratic nomination over Minnesota Senator Eugene McCarthy and South Dakota Senator George McGovern at the tumultuous 1968 Democratic National Convention. Nixon won the Republican nomination over New York Governor Nelson Rockefeller and California Governor Ronald Reagan. As of 2022, Wallace is the most recent third-party candidate to win a state's entire share of electoral votes. Nixon became the first former (non-sitting) vice president to win a presidential election; he was the only person to achieve that until former Vice President Joe Biden won the 2020 Presidential election.
The Republican Party won a net gain of five seats in both the House and the Senate. However, the Democratic Party retained strong majorities in both houses of Congress. In the gubernatorial elections, the Republican Party picked up a net gain of five governorships. This was the second consecutive election where the winning presidential party had coattails in both houses of Congress and the first for Republicans since 1952.
This was the first of two times since 1889 that a newly elected President's party failed to control either house of Congress.
|
Provide a bulleted list of the results of the 1968 United States elections.
|
The following were the results of the 1968 United States elections:
- Richard Nixon defeated Hubert Humphrey in the presidential election
- The Republican party gained five seats in both the House and Senate
- The Democratic Party maintained majority in both the House and Senate
|
null | false
| null |
The Blue Angels are a flight demonstration squadron of the United States Navy. Formed in 1946, the unit is the second oldest formal aerobatic team in the world, after the French Patrouille de France formed in 1931. The team, composed of six Navy and one Marine Corps demonstration pilot, fly Boeing F/A-18 Super Hornets.
|
Extract when the Blue Angels formed
|
The mission of the United States Navy Flight Demonstration Squadron is to showcase the pride and professionalism of the United States Navy and Marine Corps by inspiring a culture of excellence and service to the country through flight demonstrations and community outreach. The Blue Angels are a flight demonstration squadron of the United States Navy that formed in 1946
|
null | false
| null |
Name a list of animals that you would expect to find on a farm.
|
Domesticated farm animals may include cows, pigs, sheep, chicken, turkey and ducks.
|
|
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 encoder is a bidirectional Long-Short Term Memory(LSTM) cell BIBREF14 and the decoder a single LSTM cell with attention mechanism.
|
How many attention layers are there in their model?
|
one
|
null | false
| 91
|
Neural machine translation (NMT) has achieved the state-of-the-art results on a mass of language pairs with varying structural differences, such as English-French BIBREF0, BIBREF1 and Chinese-English BIBREF2. However, so far not much is known about how and why NMT works, which pose great challenges for debugging NMT models and designing optimal architectures.
The understanding of NMT models has been approached primarily from two complementary perspectives. The first thread of work aims to understand the importance of representations by analyzing the linguistic information embedded in representation vectors BIBREF3, BIBREF4 or hidden units BIBREF5, BIBREF6. Another direction focuses on understanding the importance of input words by interpreting the input-output behavior of NMT models. Previous work BIBREF7 treats NMT models as black-boxes and provides explanations that closely resemble the attention scores in NMT models. However, recent studies reveal that attention does not provide meaningful explanations since the relationship between attention scores and model output is unclear BIBREF8.
In this paper, we focus on the second thread and try to open the black-box by exploiting the gradients in NMT generation, which aims to estimate the word importance better. Specifically, we employ the integrated gradients method BIBREF9 to attribute the output to the input words with the integration of first-order derivatives. We justify the gradient-based approach via quantitative comparison with black-box methods on a couple of perturbation operations, several language pairs, and two representative model architectures, demonstrating its superiority on estimating word importance.
We analyze the linguistic behaviors of words with the importance and show its potential to improve NMT models. First, we leverage the word importance to identify input words that are under-translated by NMT models. Experimental results show that the gradient-based approach outperforms both the best black-box method and other comparative methods. Second, we analyze the linguistic roles of identified important words, and find that words of certain syntactic categories have higher importance while the categories vary across language. For example, nouns are more important for Chinese$\Rightarrow $English translation, while prepositions are more important for English-French and -Japanese translation. This finding can inspire better design principles of NMT architectures for different language pairs. For instance, a better architecture for a given language pair should consider its own language characteristics.
Second, we analyze the linguistic roles of identified important words, and find that words of certain syntactic categories have higher importance while the categories vary across language.
|
What do the authors find by analyzing the linguistic roles of identified important words?
|
Find that words of certain syntactic categories have higher importance while the categories vary across language.
|
null | false
| null |
How is the price of an Uber ride calculated?
|
The price of an Uber ride is calculated based on supply and demand while considering the time, distance, and destination of the trip.
|
|
null | false
| null |
What are steps to become a US citizen?
|
Step 1. Determine if you are already a U.S. citizen
Step 2. Determine if you are eligible to become a U.S. citizen
Step 3. Prepare your Form N-400, Application for Naturalization
Step 4. Submit your Form N-400 and pay your fees
Step 5. Go to your biometrics appointment, if applicable
Step 6. Complete the interview
Step 7. Receive a decision from USCIS on your Form N-400
Step 8. Receive a notice to take the Oath of Allegiance
Step 9. Take the Oath of Allegiance to the United States
Step 10. Understanding U.S. citizenship
|
|
null | false
| null |
Give me a one line summary of the story below:
A long long time ago the Queen of Odisha heard that there was a master stone sculptor who made such stunning statues of the gods that people would weep from happiness at the beauty of the statues if they saw them. One summer day, while resting indolently with the King in the main hall of the royal palace, the Queen suddenly had a thought. "How nice it would be, King, if we could build some beautiful statutes of Lord Jagannath so that the people can come worship Jagannath, Balabhadra, and Subadhra. What do you think?"
The King thought for a few minutes, and then nodded in agreement. "What do you have in mind, my Queen?" And so, soon the stone sculptor was summoned to appear before the Court for an appointment to build in his way the statutes of the gods. Soon the monsoon season started and the sculptor appeared in front of the court, awaiting his appointment. He was a humble man from the mountainous regions of Odisha. Even at his best-dressed, one couldn't help but notice his rough hands, and the small sprinkling of dust that seemed ever present in his hair.
The King and Queen graciously waited as the sculptor bowed and then asked "Your majesties, what can I do for you?" The Queen leaned forward in her throne and her voice rang out in the court "We have heard, o Master Sculptor, of your skills from people far and wide across the Kingdom. We dream of having you put your skills to use to build beautiful statues that will represent Lord Jagannath, Subhadra, and Balabhadra for our people to worship."
The sculptor was silent. As his silence extended, ministers of the court began to feel restless. "Speak up! How can you be silent when the Queen graces you so!" But still, the sculptor was silent.
The King and Queen sat patiently in their thrones as they waited for the sculptor to respond. Perhaps the enormity of the task was too much. Perhaps, the sculptor was thinking of what he should ask for in return. Or even more, perhaps, the sculptor was awed and honored by the request.
At last, as the shadows from the late afternoon sun began to stretch across the floors of the hall, the sculptor shifted his weight, and opened his mouth to speak. "Your majesties, you grace me with such a request, I cannot help but oblige. It is an honor." At this, the Queen sat as straight as a rod, listening for the words that would come next. "But, I can do this on one condition only. I will work in a closed hall, where none may enter - regardless of whether they can hear me working or otherwise. None may enter until I open the doors from within and proclaim to all that the statues are completed."
The ministers gasped - what a strange request from this man. Was he a fool? What if he just slept and ate and didn't do any work? How could the King and Queen trust that he would do the work he was promising to do? The Queen looked on thoughtfully. She glanced at the King, and it seemed to her that he agreed on this condition without reservation. She looked at the sculptor and then nodded her assent. "We decree that you shall build these statues in a hall set aside, into which no one shall enter until your work is complete and you proclaim that to us after opening the doors of the hall from within."
|
The Queen of Odisha invites a master stone sculptor to build statutes to the gods, Jagannath, Balabhadra, and Subhadra, with the promise that no one will disturb the sculptor until the sculptor finishes the statues and tells everyone that he has finished the statues.
|
|
2004.03744
| false
| null |
We used Amazon Mechanical Turk (MTurk) to collect new labels and explanations for SNLI-VE. 2,060 workers participated in the annotation effort, with an average of 1.98 assignments per worker and a standard deviation of 5.54. We required the workers to have a previous approval rate above 90%. No restriction was put on the workers' location.
We used Amazon Mechanical Turk (MTurk) to collect new labels and explanations for SNLI-VE. 2,060 workers participated in the annotation effort, with an average of 1.98 assignments per worker and a standard deviation of 5.54.
|
How many annotators are used to write natural language explanations to SNLI-VE-2.0?
|
The answers are shown as follows:
* 2,060 workers
|
null | false
| null |
What happened to Captain Bligh after the Mutiny on the Bounty?
|
Captain Bligh and eighteen other men were set adrift in mid-ocean by the mutineers. Navigating by memory and astronomical observation alone, Bligh led the tiny boat all the way to the Dutch colony of Timor. The men suffered greatly from hunger, thirst, and dangerous weather. After reaching Timor, they sailed a proper ship to Batavia and, from there, returned to England.
|
|
null | false
| 249
|
As pre-processing, we applied tokenisation, PoS-tagging and lemmatisation to the data using the LeTs Preprocess Toolkit BIBREF60 . In supervised learning, a machine learning algorithm takes a set of training instances (of which the label is known) and seeks to build a model that generates a desired prediction for an unseen instance. To enable the model construction, all instances are represented as a vector of features (i.e., inherent characteristics of the data) that contain information that is potentially useful to distinguish cyberbullying from non-cyberbullying content.
We experimentally tested whether cyberbullying events can be recognised automatically by lexical markers in a post. To this end, all posts were represented by a number of information sources (or features) including lexical features like bags-of-words, sentiment lexicon features and topic model features, which are described in more detail below. Prior to feature extraction, some data cleaning steps were executed, such as the replacement of hyperlinks and @-replies, removal of superfluous white spaces, and the replacement of abbreviations by their full form (based on an existing mapping dictionary ). Additionally, tokenisation was applied before INLINEFORM0 -gram extraction and sentiment lexicon matching, and stemming was applied prior to extracting topic model features.
After pre-processing of the corpus, the following feature types were extracted:
Word INLINEFORM0 -gram bag-of-words: binary features indicating the presence of word unigrams, bigrams and trigrams.
Character INLINEFORM0 -gram bag-of-words: binary features indicating the presence of character bigrams, trigrams and fourgrams (without crossing word boundaries). Character INLINEFORM1 -grams provide some abstraction from the word level and provide robustness to the spelling variation that characterises social media data.
Term lists: one binary feature derived for each one out of six lists, indicating the presence of an item from the list in a post: proper names, `allness' indicators (e.g. always, everybody), diminishers (e.g. slightly, relatively), intensifiers (e.g. absolutely, amazingly), negation words and aggressive language and profanity words. Person alternation is a binary feature indicating whether the combination of a first and second person pronoun occurs in order to capture interpersonal intent.
Subjectivity lexicon features: positive and negative opinion word ratios, as well as the overall post polarity were calculated using existing sentiment lexicons. For Dutch, we made use of the Duoman BIBREF61 and Pattern BIBREF62 lexicons. For English, we included the Hu and Liu opinion lexicon BIBREF63 , the MPQA lexicon BIBREF64 , General Inquirer Sentiment Lexicon BIBREF65 , AFINN BIBREF66 , and MSOL BIBREF67 . For both languages, we included the relative frequency of all 68 psychometric categories in the Linguistic Inquiry and Word Count (LIWC) dictionary for English BIBREF68 and Dutch BIBREF69 .
Topic model features: by making use of the Gensim topic modelling library BIBREF70 , several LDA BIBREF71 and LSI BIBREF72 topic models with varying granularity ( INLINEFORM0 = 20, 50, 100 and 200) were trained on data corresponding to each fine-grained category of a cyberbullying event (e.g. threats, defamations, insults, defenses). The topic models were based on a background corpus (EN: INLINEFORM1 tokens, NL: INLINEFORM2 tokens) scraped with the BootCAT BIBREF73 web-corpus toolkit. BootCaT collects ASKfm user profiles using lists of manually determined seed words that are characteristic of the cyberbullying categories.
When applied to the training data, this resulted in INLINEFORM0 and INLINEFORM1 features for English and Dutch, respectively.
Prior to feature extraction, some data cleaning steps were executed, such as the replacement of hyperlinks and @-replies, removal of superfluous white spaces, and the replacement of abbreviations by their full form.
|
What data cleaning setups were executed?
|
The replacement and @-replies, removal of superfluous white spaces, and the replacement of abbreviations were used.
|
null | false
| null |
Which characters belong to DC or Marvel Universe? Ultron, Brainiac
|
Brainiac is DC, Ultron is Marvel
|
|
null | false
| 417
|
Explanation Methods for DNN While there are diverse suites of methods to explain DNN, it can be categorized based on the access of the model's internal information and how they provide explanations. The model-agnostic method is to approximate the relationships between input data and output decision while treating the model as a black-box. There are methods to create a local explanation by selecting features that play an essential role in making decisions and methods providing an interpretable model through a global approximate. On the other hand, in knowing the internal information of the model, the model-transparent method provides explanations by saliency map using gradient in classifier, class activation map using global average pooling, and feature map based on attention. The third method provides explanations by examples capturing relationship between test input and the model's decision.
Considering that examples are based on training data or learned prototype data, our work is closely aligned with these prototype-based classification models. The example-based method shows its efficacy to explain complex concepts in psychology and education domains, and is also accepted by the general public as the most preferred explanation method to explain.
The following three prototype-based classification methods are similar to our method in that they use the distance between the encoded input and prototypes for classification and explanation. learn and visualize the prototypes for the entire data through autoencoder. go further from here, characterizing training image patches as class-specific prototypes for partial explanations, while feature a hierarchical demonstration of prototypes through related groups for non-flat explanations. As most of prototype-based explanation methods are based on, we use it as a baseline model. Our framework goes further from these methods, expands prototypes from vectors to distribution in latent space and constructs explanation space for diverse explanations with reliability. In addition to prototype-based methods, there are several methods that use the distance in latent space to understand the internal behavior of a model or to provide explanation using VAE or GAN.
The largest body of research on estimating uncertainty in deep learning is Bayesian neural networks.
To obtain predictive uncertainty, Bayesian neural networks use prior distributions on model parameters and infer the posterior distribution. Although exact inference in Bayesian methods is intractable, a range of approximate techniques has been proposed, such as variational inference and stochastic gradient hamiltonian monte carlo (SG-HMC). However, in practice, these Bayesian methods are outperformed by Deep Ensembles. This non-bayesian technique uses multiple neural networks trained from different initializations and averages their predictions as the model output. It is simple and effective but requires high computational costs at training and test time. Another non-bayesian technique is based on evidential deep learning, which we use to estimate uncertainty. These methods use a single deterministic neural network and apply prior distributions on model predictions, not model parameters.
As most of prototype-based explanation methods are based on Li et al. (2018), we use it as a baseline model.
|
Are baselines missing in the evaluation section?
|
In our paper, the comparison with prior work is written in section 4.1. As mentioned in the Introduction, we wanted to develop an inherently interpretable model. Among these inherently interpretable models, we wanted to create a model that provides explanations through examples that are known as powerful explanations. That's why we created a prototype-based explanation model, as shown in Section 2 Explanation Methods for DNN.
In this process, since prototype-based DNNs are mostly made based on the method of [7], there is a problem that similar latent representations do not share a similar appearance. We developed a model to solve these common problems and expand prototype-based explanations. Therefore, as shown in the second paragraph of Explanation Methods for DNN in Section 2, we compared with [7], the paper used as the basis for prototype-based DNN, as a baseline.
Therefore, comparison with the method used as the baseline was carried out in section 4.1, and comparison with methods corresponding to the post-hoc explanation was not carried out. After comparison with the prior work, it was shown that the additional explanations obtained with our method aided the understanding of the judgment of the model in section 5.
|
null | false
| null |
What are the different types of shots in ice hockey
|
-Slapshot
-Wrist shot
-Backhand
-Snapshot
|
|
null | false
| null |
Why should I move to New York City?
|
New York, New York, also known as New York City or NYC, is the largest city in the United States. Comprised of 5 boroughs (Manhattan, Staten Island, Brooklyn, Bronx, Queens), each neighborhood has its own distinct culture and purpose that contributes to the energy of the city. Being that it is such a large city, there is something for everyone. It has an incredibly diverse population, a culture of historical culinary tradition with new cuisines arriving every day, a strongly developed art scene for film-lovers, musicians, visual artists, and so much more. There are so many reasons why it is considered a global city. It is incredibly walkable and public transit-friendly, and the green spaces alleviate the concrete jungle that New York is so well-known for. You will meet all types of people here. While the people may be controversially and arguably known for being unfriendly, they are surely kind. It is truly an unmatched experience to live, know, and love the city of New York.
|
|
null | false
| 323
|
Major progress has been made in different tasks in Natural Language Processing, yet our models are still not able to describe why they make their decisions when summarizing an article, translating a sentence, or answering a question. Lack of meta information (e.g. general world knowledge regarding the task) is one important obstacle in the construction of language understanding models capable of reasoning about their considerations when making decisions (predictions).
Wikipedia is a great resource of world knowledge for human beings, but lacks the proper structure to be useful for the models. To address this issue and make a more structured knowledge-base, we are trying to structure Wikipedia. The final goal is to have, for each Wikipedia article, known entities and sets of attributes, with each attribute linking to other entities wherever possible. The initial step towards this goal is to classify the entities into predefined categories and verify the results using human annotators.
Throughout the past years, many have tried classifying Wikipedia articles into different category sets the majority of which range between 3 to 15 class types BIBREF0, BIBREF1, BIBREF2, BIBREF3, BIBREF4. Although useful, such categorization type sets are not much helpful when the classified articles are being used as the training data for Question-Answering systems, since the extracted knowledge-base does not provide detailed enough information to the model.
On the other hand, much larger categorization type sets such as Cyc-Taxonomy BIBREF5, Yago-Taxonomy BIBREF6, or Wikipedia's own taxonomy of categories BIBREF7 are not suitable for our task since the tags are not verifiable for annotators. In addition, taxonomies are not designed in a tree format, so some categories might have multiple super-categories and this would make the verification process much harder for the cases that the article is about multiple different topics.
Considering the mentioned problem requirements, we believe Extended Named Entities Hierarchy BIBREF8, containing 200 fine-grained categories tailored for Wikipedia articles, is the best fitting tag set.
Higashinaka et al. higashinaka2012 were the first to use this extended tag set as the categorization output of the dumped Wikipedia pages while using a hand-extracted feature set for converting the pages into their model input vectors. Following their work, Suzuki et al. suzuki2016 modelled the links between different Wikipedia pages as an augmentation to the extracted input features to the classifier. They also proposed a more complex model for learning the mapping between the converted articles and the labels.
Although providing useful insights, none of the works above have considered the multi-lingual nature of many Wikipedia articles. Hence, we decided to hire annotators and educate them on the Extended Named Entities (ENE) tag set to annotate each article with up to 6 different ENE classes, and exploit the Wikipedia language links in the annotated articles to create our multi-lingual Wikipedia classification dataset. Section 2 details our dataset creation process.
We then used the two models mentioned above, which are to the best of our knowledge the only works close enough to our task in hand, to benchmark our dataset. Section 3 provides more details about our feature selection method and the models. Section 4 presents our experimental setup and the classification results.
Although providing useful insights, none of the works above have considered the multi-lingual nature of many Wikipedia articles. Hence, we decided to hire annotators and educate them on the Extended Named Entities (ENE) tag set to annotate each article with up to 6 different ENE classes, and exploit the Wikipedia language links in the annotated articles to create our multi-lingual Wikipedia classification dataset.
|
Why did the authors decide to hire annotators and educate them on the Extended Named Entities (ENE) tag set?
|
Because none of the works above have considered the multi-lingual nature of many Wikipedia articles and they want to annotate each article with up to 6 different ENE classes, and exploit the Wikipedia language links in the annotated articles to create a multi-lingual Wikipedia classification dataset.
|
1909.08859
| false
| null |
Table TABREF29 presents the quantitative results for the visual reasoning tasks in RecipeQA. In single-task training setting, PRN gives state-of-the-art results compared to other neural models. Moreover, it achieves the best performance on average. These results demonstrate the importance of having a dynamic memory and keeping track of entities extracted from the recipe. In multi-task training setting where a single model is trained to solve all the tasks at once, PRN and BIDAF w/ static memory perform comparably and give much better results than BIDAF. Note that the model performances in the multi-task training setting are worse than single-task performances. We believe that this is due to the nature of the tasks that some are more difficult than the others. We think that the performance could be improved by employing a carefully selected curriculum strategy BIBREF20.
FLOAT SELECTED: Table 1: Quantitative comparison of the proposed PRN model against the baselines.
Table TABREF29 presents the quantitative results for the visual reasoning tasks in RecipeQA. In single-task training setting, PRN gives state-of-the-art results compared to other neural models.
In multi-task training setting where a single model is trained to solve all the tasks at once, PRN and BIDAF w/ static memory perform comparably and give much better results than BIDAF.
FLOAT SELECTED: Table 1: Quantitative comparison of the proposed PRN model against the baselines.
|
How better is accuracy of new model compared to previously reported models?
|
Average accuracy of proposed model vs best prevous result:
Single-task Training: 57.57 vs 55.06
Multi-task Training: 50.17 vs 50.59
|
null | false
| null |
Most transistors are made from very pure silicon, and some from germanium, but certain other semiconductor materials are sometimes used. A transistor may have only one kind of charge carrier, in a field-effect transistor, or may have two kinds of charge carriers in bipolar junction transistor devices. Compared with the vacuum tube, transistors are generally smaller and require less power to operate. Certain vacuum tubes have advantages over transistors at very high operating frequencies or high operating voltages. Many types of transistors are made to standardized specifications by multiple manufacturers.
|
What are transistors made from?
|
silicon, germanium
|
null | false
| null |
Sensory processing disorder (SPD, formerly known as sensory integration dysfunction) is a condition in which multisensory input is not adequately processed in order to provide appropriate responses to the demands of the environment. Sensory processing disorder is present in many people with autism spectrum disorder and attention deficit hyperactivity disorder. Individuals with SPD may inadequately process visual, auditory, olfactory (smell), gustatory (taste), tactile (touch), vestibular (balance), proprioception (body awareness), and interoception (internal body senses) sensory stimuli.
Sensory integration was defined by occupational therapist Anna Jean Ayres in 1972 as "the neurological process that organizes sensation from one's own body and from the environment and makes it possible to use the body effectively within the environment".Sensory processing disorder has been characterized as the source of significant problems in organizing sensation coming from the body and the environment and is manifested by difficulties in the performance in one or more of the main areas of life: productivity, leisure and play or activities of daily living.
Sources debate whether SPD is an independent disorder or represents the observed symptoms of various other, more well-established, disorders.SPD is not included in the Diagnostic and Statistical Manual of Mental Disorders of the American Psychiatric Association, and the American Academy of Pediatrics has recommended in 2012 that pediatricians not use SPD as a stand-alone diagnosis.
|
What is Sensory processing disorder?
|
Sensory processing disorder (SPD, formerly known as sensory integration dysfunction) is a condition in which multisensory input is not adequately processed to provide appropriate responses to the demands of the environment. Sensory processing disorder is present in many people with autism spectrum disorder and attention deficit hyperactivity disorder. Individuals with SPD may inadequately process visual, auditory, olfactory (smell), gustatory (taste), tactile (touch), vestibular (balance), proprioception (body awareness), and interoception (internal body senses) sensory stimuli.
|
null | false
| null |
What is the meaning of life and how do I have a fulfilling one?
|
The meaning of life is to live a long and fulfilling one. A good life may comprise of having many friends, meaningful relationships, a family, kids, and being a constructive member of society through hard work and positive interactions. It is always best to leave a positive impression on those you meet or who you come into contact with during your daily routine. You will want to focus and put positive intention in all of your actions so that you leave a good impression on all those who come into contact with you. Don't forget that part of life is enjoying experiences and having fun too. Find hobbies you enjoy, and do them often. We say this is important because it will lead to never ending happiness. As you age you will get wiser, and it will also be important to impart wisdom on those individuals who are younger than you who still need to experience life in order to gain the same wisdom.
|
|
null | false
| null |
Classify each of the following as a conductor or insulator: steel, copper, rubber, wood, cardboard, aluminum
|
Steel: Conductor
Copper: Conductor
Rubber: Insulator
Wood: Insulator
Cardboard: Insulator
Aluminum: Conductor
|
|
null | false
| 50
|
Equations are an important part of scientific articles, but many existing machine learning methods do not easily handle them. They are challenging to work with because each is unique or nearly unique; most equations occur only once. An automatic understanding of equations, however, would significantly benefit methods for analyzing scientific literature. Useful representations of equations can help draw connections between articles, improve retrieval of scientific texts, and help create tools for exploring and navigating scientific literature.
In this paper we propose equation embeddings (EqEmb), an unsupervised approach for learning distributed representations of equations. The idea is to treat the equation as a "singleton word," one that appears once but that appears in the context of other words. The surrounding text of the equation—and in particular, the distributed representations of that text—provides the data we need to develop a useful representation of the equation.
Figure FIGREF1 illustrates our approach. On the left is an article snippet BIBREF0 . Highlighted in orange is an equation; in this example it represents a neural network layer. We note that this particular equation (in this form and with this notation) only occurs once in the collection of articles (from arXiv). The representations of the surrounding text, however, provide a meaningful context for the equation. Those words allow us to learn its embedding, specifically as a "word" which appears in the context of its surroundings. The resulting representation, when compared to other equations' representations and word representations, helps find both related equations and related words. These are illustrated on the right.
EqEmbs build on exponential family embeddings BIBREF1 to include equations as singleton observations and to model equation elements such as variables, symbols and operators. Exponential family embeddings, like all embedding methods, define a context of each word. In our initial EqEmb, the context for the words is a small window, such as four or eight words, but the context of an equation is a larger window, such as sixteen words. Using these two types of contexts together finds meaningful representations of words and equations. In the next EqEmb, which builds on the first, we consider equations to be sentences consisting of equation units, i.e., variables, symbols, and operators. Equation units help model equations across two types of context—over the surrounding units and over the surrounding words.
We studied EqEmbs on four collections of scientific articles from the arXiv, covering four computer science domains: natural language processing (NLP), information retrieval (IR), artificial intelligence (AI) and machine learning (ML). We found that EqEmbs provide more efficient modeling than existing word embedding methods. We further carried out an exploratory analysis of a large set of INLINEFORM0 87k equations. We found that EqEmbs provide better models when compared to existing word embedding approaches. EqEmbs also provide coherent semantic representations of equations and can capture semantic similarity to other equations and to words.
EqEmbs also provide coherent semantic representations of equations and can capture semantic similarity to other equations and to words.
|
What about the performance of EqEmbs on coherent semantic representations of equations?
|
EqEmbs also provide coherent semantic representations of equations and can capture semantic similarity to other equations and to words.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.