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 |
Which is a species of fish? Mullet or Afro
|
Mullet
|
|
null | false
| null |
What type of pokemon should I use to battle a water pokemon? Give me some examples.
|
Water type pokemon are weak against grass and electric type pokemon. Therefore, you should choose one of your pokemon of grass or electric type. The best choice, if you want to win the immediate battle, is to pick your strongest pokemon of the grass or electric type. Some examples of strong grass or electric type pokemon are Raichu, Venosaur, or Zapados.
|
|
null | false
| null |
As of 2013, UVA's $1.4 billion academic budget is paid for primarily by tuition and fees (32%), research grants (23%), endowment and gifts (19%), and sales and services (12%). The university receives 10% of its academic funds through state appropriation from the Commonwealth of Virginia. For the overall (including non-academic) university budget of $2.6 billion, 45% comes from medical patient revenue. The Commonwealth contributes less than 6%.
|
According to this paragraph about UVA, tell me how much of UVA's academic budget foes to endowment and gifts?
|
19% of UVA's academic budget goes to endowment and gifts.
|
null | false
| null |
Can I stream data directly into and from Delta tables?
|
Yes, you can use Structured Streaming to directly write data into Delta tables and read from Delta tables. See Stream data into Delta tables and Stream data from Delta tables.
|
|
null | false
| null |
List some of the best life changing self-help books
|
Atomic Habits by James Clear
Power of discipline by Daniel Walter
The Power of Habit by Charles Duhigg
Tiny Habits by BJ Fogg
High Performance Habits by Brendon Burchard
Getting Things Done by David Allen
What the Most Successful People Do Before Breakfast by Laura Vanderkam
How to Change by Katy Milkman
Start With Why by Simon Sinek
The Culture Code by Daniel Coyle
|
|
1610.09516
| false
| null |
It has been recognized that music is a key cultural component in an urban lifestyle and that gang members often want to emulate the scenarios and activities the music conveys BIBREF7 . Our analysis confirms that the influence of gangster rap is expressed in gang members' Twitter posts. We found that 51.25% of the gang members collected have a tweet that links to a YouTube video. Following these links, a simple keyword search for the terms gangsta and hip-hop in the YouTube video description found that 76.58% of the shared links are related to hip-hop music, gangster rap, and the culture that surrounds this music genre. Moreover, this high proportion is not driven by a small number of profiles that prolifically share YouTube links; eight YouTube links are shared on average by a gang member.
We found that 51.25% of the gang members collected have a tweet that links to a YouTube video. Following these links, a simple keyword search for the terms gangsta and hip-hop in the YouTube video description found that 76.58% of the shared links are related to hip-hop music, gangster rap, and the culture that surrounds this music genre.
|
What are the differences in the use of YouTube links between gang member and the rest of the Twitter population?
|
The answers are shown as follows:
* 76.58% of the shared links are related to hip-hop music, gangster rap, and the culture that surrounds this music genre
|
1910.10869
| false
| null |
In spite of the windowing approach, the class distribution is still skewed, and an accuracy metric would reflect the particular class distribution in our data set. Therefore, we adopt the unweighted average recall (UAR) metric commonly used in emotion classification research. UAR is a reweighted accuracy where the samples of both classes are weighted equally in aggregate. UAR thus simulates a uniform class distribution. To match the objective, our classifiers are trained on appropriately weighted training data. Note that chance performance for UAR is by definition 50%, making results more comparable across different data sets.
In spite of the windowing approach, the class distribution is still skewed, and an accuracy metric would reflect the particular class distribution in our data set. Therefore, we adopt the unweighted average recall (UAR) metric commonly used in emotion classification research. UAR is a reweighted accuracy where the samples of both classes are weighted equally in aggregate. UAR thus simulates a uniform class distribution. To match the objective, our classifiers are trained on appropriately weighted training data. Note that chance performance for UAR is by definition 50%, making results more comparable across different data sets.
|
What they use as a metric of finding hot spots in meeting?
|
The answers are shown as follows:
* unweighted average recall (UAR) metric
|
null | false
| null |
Gardening is the practice of growing and cultivating plants as part of horticulture. In gardens, ornamental plants are often grown for their flowers, foliage, or overall appearance; useful plants, such as root vegetables, leaf vegetables, fruits, and herbs, are grown for consumption, for use as dyes, or for medicinal or cosmetic use.
Gardening ranges in scale from fruit orchards, to long boulevard plantings with one or more different types of shrubs, trees, and herbaceous plants, to residential back gardens including lawns and foundation plantings, all the way to container gardens grown inside or outside. Gardening may be very specialized, with only one type of plant grown, or involve a variety of plants in mixed plantings. It involves an active participation in the growing of plants, and tends to be labor-intensive, which differentiates it from farming or forestry.
|
What is gardening?
|
Gardening is laying out and caring for a plot of ground devoted partially or wholly to growing plants such as flowers, herbs, or vegetables.
|
null | false
| null |
What is Hornachuelos?
|
Hornachuelos is a small, picturesque village in the Southern Region of Spain called Andalusia. It is famous for its ancient history, Roman walls and Natural Park.
|
|
null | false
| null |
What are the interests behind climate change denial?
|
Financial interest from industries opposed to the regulation of CO2 emissions. Migrating to low carbon technologies would require massive investments, making executives concerned about impacts on companies gross margin.
|
|
null | false
| null |
What are the best cliffs in Ireland?
|
The Cliffs of Moher are 702 feet/214 meters at their highest point, near O'Brien's Tower. Though not the highest cliffs in Ireland – that honor goes to Co Donegal's Slieve League Cliffs and Achill Island's Croaghaun Cliffs – they are by far the most popular.
|
|
null | false
| null |
Classify each of the following as fluffy or hard: lamb, sheep, poodle, cement, stuffed animal, golf ball
|
Lamb: Fluffy
Sheep: Fluffy
Poodle: Fluffy
Cement: Hard
Stuffed Animal: Fluffy
Golf Ball: Hard
|
|
null | false
| null |
Marketing Automation is a subset of customer relationship management (CRM) or customer experience management (CXM) that focuses on the definition, segmentation, scheduling and tracking of marketing campaigns. The use of marketing automation makes processes that would otherwise have been performed manually much more efficient and makes new processes possible. Marketing Automation can be defined as a process where technology is used to automate several repetitive tasks that are undertaken on a regular basis in a marketing campaign.
Marketing Automation platforms allow marketers to automate and simplify client communication by managing complex omnichannel marketing strategies from a single tool. Marketing Automation assists greatly in areas like Lead Generation, Segmentation, Lead nurturing and lead scoring, Relationship marketing, Cross-sell and upsell, Retention, Marketing ROI measurement. Effective marketing automation tools leverage data from a separate or integrated CRM to understand customer impact and preferences.
|
What is Marketing Automation?
|
Marketing Automation is software that helps marketers to do create efficient online and offline campaigns. It's part of CRM suite and people usually integrate it with their salesforce application and marketing website. The Marketing Automation platform allows marketers to automate and simplify customer communication by managing multi communication channels. This includes personalized email, dynamic web pages, digital banners, social media and more. People use it for lead generation, segmentation and nurturing and lead scoring. Post-sales marketers also use Marketing Automation for cross-sell, upsell and retention marketing. In
|
null | false
| 37
|
With this work we present a resource that will be extremely useful for building language systems in an endangered, under-represented language, Mapudungun. We benchmark NLP systems for speech synthesis, speech recognition, and machine translation, providing strong baseline results. The size of our resource (142 hours, more than 260k total sentences) has the potential to alleviate many of the issues faced when building language technologies for Mapudungun, in contrast to other indigenous languages of the Americas that unfortunately remain low-resource.
Our resource could also be used for ethnographic and anthropological research into the Mapuche culture, and has the potential to contribute to intercultural bilingual education, preservation activities and further general advancement of the Mapudungun-speaking community.
We benchmark NLP systems for speech synthesis, speech recognition, and machine translation, providing strong baseline results.
|
What do the authors benchmark NLP systems for?
|
The authors benchmark NLP systems for speech synthesis, speech recognition, and machine translation, providing strong baseline results.
|
1605.08675
| false
| null |
Secondly, texts go through a cascade of annotation tools, enriching it with the following information:
Morphosyntactic interpretations (sets of tags), using Morfeusz 0.82 BIBREF25 ,
Tagging (selection of the most probable interpretation), using a transformation-based learning tagger, PANTERA 0.9.1 BIBREF26 ,
Syntactic groups (possibly nested) with syntactic and semantic heads, using a rule-based shallow parser Spejd 1.3.7 BIBREF27 with a Polish grammar, including improved version of modifications by BIBREF28 , enabling lemmatisation of nominal syntactic groups,
Named entities, using two available tools: NERF 0.1 BIBREF29 and Liner2 2.3 BIBREF30 .
Secondly, texts go through a cascade of annotation tools, enriching it with the following information:
Morphosyntactic interpretations (sets of tags), using Morfeusz 0.82 BIBREF25 ,
Tagging (selection of the most probable interpretation), using a transformation-based learning tagger, PANTERA 0.9.1 BIBREF26 ,
Syntactic groups (possibly nested) with syntactic and semantic heads, using a rule-based shallow parser Spejd 1.3.7 BIBREF27 with a Polish grammar, including improved version of modifications by BIBREF28 , enabling lemmatisation of nominal syntactic groups,
Named entities, using two available tools: NERF 0.1 BIBREF29 and Liner2 2.3 BIBREF30 .
|
How is the data in RAFAEL labelled?
|
Using a set of annotation tools such as Morfeusz, PANTERA, Spejd, NERF and Liner
|
null | false
| null |
What is a graph?
|
Graph is a set on nodes that are connected via edges. Edges represent the relationships between the nodes and can have multiple properties like weight or sign. For example the weight of the edge can represent how strong are the relationships between the nodes A and B, while the sign might represent wether these relationships are positive or negative.
|
|
1612.05310
| true
| null |
The overall Total Accuracy score reported in table TABREF19 using the entire feature set is 549. This result is what makes this dataset interesting: there is still lots of room for research on this task. Again, the primary goal of this experiment is to help identify the difficult-to-classify instances for analysis in the next section.
The overall Total Accuracy score reported in table TABREF19 using the entire feature set is 549.
|
Do they experiment with the dataset?
|
Yes.
|
null | false
| 170
|
We applied this method to discover gender-associated words in both domains. In Table TABREF9, we present a sample of the most gender-associated nouns from the celebrity domain. Several themes emerge: for example, female celebrities seem to be more associated with appearance (“gown,” “photo,” “hair,” “look”), while male celebrities are more associated with creating content (“movie,” “film,” “host,” “director”). This echoes real-world trends: for instance, on the red carpet, actresses tend to be asked more questions about their appearance –- what brands they are wearing, how long it took to get ready, etc. –- while actors are asked questions about their careers and creative processes (as an example, see BIBREF31).
Table TABREF9 also includes some of the most gender-associated verbs and adjectives from the professor domain. Female CS professors seem to be praised for being communicative and personal with students (“respond,” “communicate,” “kind,” “caring”), while male CS professors are recognized for being knowledgeable and challenging the students (“teach,”, “challenge,” “brilliant,” “practical”). These trends are well-supported by social science literature, which has found that female teachers are praised for “personalizing” instruction and interacting extensively with students, while male teachers are praised for using “teacher as expert” styles that showcase mastery of material BIBREF32.
These findings establish that there are clear differences in how people talk about women and men – even with Bonferroni correction, there are still over 500 significantly gender-associated nouns, verbs, and adjectives in the celebrity domain and over 200 in the professor domain. Furthermore, the results in both domains align with prior studies and real world trends, which validates that our methods can capture meaningful patterns and innovatively provide evidence at the large-scale. This analysis also hints that it can be helpful to abstract from words to topics to recognize higher-level patterns of gender associations, which motivates our next section on clustering.
Table TABREF11 displays a sample of our results – we find that the clusters are coherent in context and the labels seem reasonable. In the next section, we discuss human evaluations that we conducted to more rigorously evaluate the output, but first we discuss the value of these methods toward analysis.
At the word-level, we hypothesized that in the celebrity domain, women were more associated with appearance and men with creating content. Now, we can validate those hypotheses against labeled clusters – indeed, there is a cluster labeled clothing that is 100% female (i.e. 100% words are female-associated), and a 80% male cluster labeled movie. Likewise, in the professor domain, we had guessed that women are associated with communication and men with knowledge, and there is a 100% female cluster labeled communication and a 89% male cluster labeled cognition. Thus, cluster labeling proves to be very effective at pulling out the patterns that we believed we saw at the word-level, but could not formally validate.
The clusters we mentioned so far all lean heavily toward one gender association or the other, but some clusters are interesting precisely because they do not lean heavily – this allows us to see where semantic groupings do not align exactly with gender association. For example, in the celebrity domain, there is a cluster labeled lover that has a mix of female-associated words (“boyfriend,” “beau,” “hubby”) and male-associated words (“wife,” “girlfriend”). Jointly leveraging cluster labels and gender associations allows us to see that in the semantic context of having a lover, women are typically associated with male figures and men with female figures, which reflects heteronormativity in society.
These findings establish that there are clear differences in how people talk about women and men – even with Bonferroni correction, there are still over 500 significantly gender-associated nouns, verbs, and adjectives in the celebrity domain and over 200 in the professor domain.
|
Do the people talk about women and men in different ways?
|
Yes, they do.
|
null | false
| null |
Who are you?
|
I am a generative model trained on large corpus of natural language text.
|
|
1908.10275
| false
| null |
time: decade (classes between 1900s and 2010s) and year representative of the time when the genre became meainstream
time: decade (classes between 1900s and 2010s) and year representative of the time when the genre became meainstream
|
Which decades did they look at?
|
The answers are shown as follows:
* between 1900s and 2010s
|
null | false
| null |
The Cloquet fire was an immense forest fire in northern Minnesota, United States in October 1918, caused by sparks on the local railroads and dry conditions. The fire left much of western Carlton County devastated, mostly affecting Moose Lake, Cloquet, and Kettle River. Cloquet was hit the hardest by the fires. It was the worst natural disaster in Minnesota history in terms of the number of casualties in a single day. In total, 453 people died and 52,000 people were injured or displaced, 38 communities were destroyed, 250,000 acres (100,000 ha) were burned, and $73 million ($1.315 billion in 2023 United States dollars) in property damage was suffered. Thirteen million dollars in federal aid were disbursed.
|
How many people died in the Cloquet fire of Illinois according to the text?
|
453 people died in the Cloquet fire that occurred in northern Minnesota, United States in October 1918.
In addition to the 453 people that died, 52,000 people were injured or displaced, 38 communities were destroyed, 250,000 acres (100,000 ha) were burned, and $73 million ($1.315 billion in 2023 United States dollars) in property damage was suffered.
|
null | false
| null |
How do I cook an egg over easy?
|
To cook an egg over easy you will need a pan, stove, oil (or butter), an egg, and a spatula. Begin by placing the pan on the stove with a minimal amount of oil or butter. Heat the pan on the stove until it is hot. What’s the pan is hot crack the egg in the middle of the pan and let it cook until the egg white becomes solid. At which point flip the egg and let it cook until the desired consistency of the yoke. Finally remove from the pan and enjoy!
|
|
null | false
| 28
|
PARENT evaluates each instance INLINEFORM0 separately, by computing the precision and recall of INLINEFORM1 against both INLINEFORM2 and INLINEFORM3 .
We compute recall against both the reference (Er(Ri)), to ensure proper sentence structure in the generated text, and the table (Er(Ti)), to ensure that texts which mention more information from the table get higher scores (e.g. candidate 3 in Figure 1).
|
Why does the team compute recall against both the reference (Er(Ri)) and the table (Er(Ti))?
|
In order to ensure proper sentence structure in the generated text, as well as to ensure that texts which mention more information from the table get higher scores.
|
null | false
| null |
What is Grafana?
|
Grafana is a multi-platform open source analytics and interactive visualization web application. It provides charts, graphs, and alerts for the web when connected to supported data sources.
|
|
null | false
| null |
How many Kangaroos are in Australia?
|
There are twice as many kangaroos in Australia as there are people. The kangaroo population is estimated at about 40 million.
|
|
null | false
| null |
What is basketball?
|
Basketball is a sport where two teams play against each other on a court with two hoops. One hoop is for each team. Each teams scores a point by putting a ball through the opponent's hoop. They play against each other in a rectangular style court and each team consists of 5 players each.
|
|
2003.01769
| false
| null |
Several recent works have investigated jointly training the acoustic model with a masking speech enhancement model BIBREF11, BIBREF12, BIBREF13, but these works did not evaluate their system on speech enhancement metrics. Indeed, our internal experiments show that without access to the clean data, joint training severely harms performance on these metrics.
Several recent works have investigated jointly training the acoustic model with a masking speech enhancement model BIBREF11, BIBREF12, BIBREF13, but these works did not evaluate their system on speech enhancement metrics. Indeed, our internal experiments show that without access to the clean data, joint training severely harms performance on these metrics.
|
Which frozen acoustic model do they use?
|
The answers are shown as follows:
* a masking speech enhancement model BIBREF11, BIBREF12, BIBREF13
|
2002.01359
| false
| null |
Our data collection setup uses a dialogue simulator to generate dialogue outlines first and then paraphrase them to obtain natural utterances. Using a dialogue simulator offers us multiple advantages. First, it ensures the coverage of a large variety of dialogue flows by filtering out similar flows in the simulation phase, thus creating a much diverse dataset. Second, simulated dialogues do not require manual annotation, as opposed to a Wizard-of-Oz setup BIBREF17, which is a common approach utilized in other datasets BIBREF0. It has been shown that such datasets suffer from substantial annotation errors BIBREF18. Thirdly, using a simulator greatly simplifies the data collection task and instructions as only paraphrasing is needed to achieve a natural dialogue. This is particularly important for creating a large dataset spanning multiple domains.
Our data collection setup uses a dialogue simulator to generate dialogue outlines first and then paraphrase them to obtain natural utterances.
|
Where is the dataset from?
|
The answers are shown as follows:
* dialogue simulator
|
null | false
| 288
|
When there are a very large number of documents that need to be read in limited time, we often resort to reading summaries instead of the whole document. Automatically generating (abstractive) summaries is a problem with various applications, e.g., automatic authoring BIBREF0. We have developed automatic text summarization systems that condense large documents into short and readable summaries. It can be used for both single (e.g., BIBREF1, BIBREF2 and BIBREF3) and multi-document summarization (e.g.,BIBREF4, BIBREF3, BIBREF5).
Text summarization is broadly classified into two categories: extractive (e.g., BIBREF3 and BIBREF6) and abstractive summarization (e.g., BIBREF7, BIBREF8 and BIBREF9). Extractive approaches select sentences from a given document and groups them to form concise summaries. By contrast, abstractive approaches generate human-readable summaries that primarily capture the semantics of input documents and contain rephrased key content. The former task falls under the classification paradigm, and the latter belongs to the generative modeling paradigm, and therefore, it is a much harder problem to solve. The backbone of state-of-the-art summarization models is a typical encoder-decoder BIBREF10 architecture that has proved to be effective for various sequential modeling tasks such as machine translation, sentiment analysis, and natural language generation. It contains an encoder that maps the raw input word vector representations to a latent vector. Then, the decoder usually equipped with a variant of the attention mechanism BIBREF11 uses the latent vectors to generate the output sequence, which is the summary in our case. These models are trained in a supervised learning setting where we minimize the cross-entropy loss between the predicted and the target summary. Encoder-decoder models have proved effective for short sequence tasks such as machine translation where the length of a sequence is less than 120 tokens. However, in text summarization, the length of the sequences vary from 400 to 800 tokens, and modeling long-term dependencies becomes increasingly difficult.
Despite the metric's known drawbacks, text summarization models are evaluated using ROUGE BIBREF12, a discrete similarity score between predicted and target summaries based on 1-gram, 2-gram, and n-gram overlap. Cross-entropy loss would be a convenient objective on which to train the model since ROUGE is not differentiable, but doing so would create a mismatch between metrics used for training and evaluation. Though a particular summary scores well on ROUGE evaluation comparable to the target summary, it will be assigned lower probability by a supervised model. To tackle this problem, we have used a self-critic policy gradient method BIBREF13 to train the models directly using the ROUGE score as a reward. In this paper, we propose an architecture that addresses the issues discussed above.
To tackle this problem, we have used a self-critic policy gradient method (Rennie et al., 2016) to train the models directly using the ROUGE score as a reward.
|
What method do they use to train the models?
|
A self-critic policy gradient method.
|
1710.10609
| true
| null |
In order to identify frequent user intents, one can use existing clustering algorithms to group together all the utterances from the users. Here each cluster would correspond to a new intent and each utterance in the cluster would correspond to an example for the intent. Similarly the agents utterances can be clustered to identify system responses. However, we argue that rather than treating user utterances and agents responses in an isolated manner, there is merit in jointly clustering them. There is adjacency information of these utterances that can be utilized to identify better user intents and system responses. As an example, consider agent utterances A.2 in box A and A.2 in box C in Figure FIGREF5 (a). The utterances “Which operating system do you use?" and “What OS is installed in your machine" have no syntactic similarity and therefore may not be grouped together. However the fact that these utterances are adjacent to the similar user utterances “I am unable to start notes email client" and “Unable to start my email client" provides some evidence that the agent utterances might be similar. Similarly the user utterances “My system keeps getting rebooted" and “Machine is booting time and again" ( box B and D in Figure FIGREF5 (a))- that are syntactically not similar - could be grouped together since the adjacent agent utterances, “Is your machine heating up?" and “Is the machine heating?" are similar.
Similarly the agents utterances can be clustered to identify system responses. However, we argue that rather than treating user utterances and agents responses in an isolated manner, there is merit in jointly clustering them. There is adjacency information of these utterances that can be utilized to identify better user intents and system responses.
|
Do they study frequent user responses to help automate modelling of those?
|
Yes.
|
null | false
| null |
In Thailand before 1888 there were no permanent, public hospitals to provide care to sick people. Temporary hospitals were set up to care for patients during epidemics, then disbanded when the epidemic subsided. Under King Chulalongkorn (Rama V) a hospital was constructed and completed in 1888 and named "Siriraj Hospital" in commemoration of the king's young son, Prince Siriraj Kakudhabhand, who had died of dysentery. King Vajiravudh, King Chulalongkorn's successor, established Department of Health on 27 November 1918.
During the reign of King Rama VIII, the Ministry of Public Health was established on 10 March 1942 as a result of the enactment of the Ministries and Departments Reorganization Act (Amendment No. 3) of B.E. 2485. Later in 1966, the date 27 November was chosen as the commemoration day of the Ministry of Public Health's foundation.
|
Write some points about Ministry of Public Health (Thailand) using given text as a reference
|
1. There were no permanent public hospitals in Thailand prior to 1888 to care for the ill.
2. When an epidemic broke out, temporary hospitals were established to provide medical care for patients.
3. A hospital was built and finished in 1888 under King Chulalongkorn (Ram V), and given the name "Siriraj Hospital" in honour of the king's little son, Prince Siriraj Kakudhabhand, who had passed away from dysentery.
4. On November 27, 1918, King Vajiravudh, who succeeded King Chulalongkorn, created the Department of Health.
5. On March 10, 1942, under King Rama VIII, the Ministries and Departments Reorganization Act (Amendment No. 3) of B.E. 2485 was passed, which led to the creation of the Ministry of Public Health.
6. Later in 1966, the establishment of the Ministry of Public Health was commemorated on November 27.
|
null | false
| null |
Identify which instrument is string or percussion: Whip, Waldzither
|
Waldzither is string, Whip is percussion.
|
|
1907.11062
| false
| null |
We have decided to focus on only one specific type of job: sales positions. After filtering based on specific job titles from the ROME Database, a list of positions was selected and verified by the authors and an expert from the Human Resources (HR). Finally, in a collaboration with an HR industry actor, we have obtained a dataset of French video interviews comprising more than 475 positions and 7938 candidates. As they watch candidates' videos, recruiters can like, dislike, shortlist candidates, evaluate them on predefined criteria, or write comments. To simplify the task, we set up a binary classification: candidates who have been liked or shortlisted are considered part of the hirable class and others part of the not hirable class. If multiple annotators have annotated the same candidates, we proceed with a majority vote. In case of a draw, the candidate is considered hirable. It is important to note that the videos are quite different from what could be produced in a laboratory setup. Videos can be recorded from a webcam, a smartphone or a tablet., meaning noisy environments and low quality equipment are par for the course. Due to these real conditions, feature extraction may fail for a single modality during a candidate's entire answer. One example is the detection of action units when the image has lighting problems. We decided to use all samples available in each modality separately. Some statistics about the dataset are available in Table TABREF33 . Although the candidates agreed to the use of their interviews, the dataset will not be released to public outside of the scope of this study due to the videos being personal data subject to high privacy constraints.
To simplify the task, we set up a binary classification: candidates who have been liked or shortlisted are considered part of the hirable class and others part of the not hirable class.
|
How is "hirability" defined?
|
The answers are shown as follows:
* candidates who have been liked or shortlisted are considered part of the hirable class
|
null | false
| null |
Write an email to the city of Berkeley requesting a refund for basketball camp because it was boring.
|
To Whom It May Concern,
We signed our son up for basketball camp over the break assuming that it would be similar to the quality of the other city programs, including the basketball programs. However, our son did not want to return to camp because it felt more like military training than the program you described on the parks and recreation web page.
On your site, you describe it as a fun opportunity for children to make friends and improve their basketball skills. The coach, however, told the children that they were not allowed to speak to each other all day, which made it impossible to either have fun to make friends.
As a result, we have removed our son from the camp and are requesting a refund.
Sincerely,
[insert name]
|
|
null | false
| null |
Mumbai is the financial, commercial, and the entertainment capital of India. It is also one of the world's top ten centres of commerce in terms of global financial flow, generating 6.16% of India's GDP, and accounting for 25% of industrial output, 70% of maritime trade in India (Mumbai Port Trust and JNPT), and 70% of capital transactions to India's economy. The city houses important financial institutions and the corporate headquarters of numerous Indian companies and multinational corporations. It is also home to some of India's premier scientific and nuclear institutes. The city is also home to Bollywood and Marathi cinema industries. Mumbai's business opportunities attract migrants from all over India.
|
From the passage provided, extract which cinema industries are located in Mumbai
|
Mumbai is home to Bollywood and Marathi cinema industries.
|
null | false
| null |
Identify which instrument is string or percussion: Dhimay, Swarabat
|
Swarabat is string, Dhimay is percussion.
|
|
null | false
| 32
|
Propaganda aims at influencing people's mindset with the purpose of advancing a specific agenda. In the Internet era, thanks to the mechanism of sharing in social networks, propaganda campaigns have the potential of reaching very large audiences BIBREF0, BIBREF1, BIBREF2.
Propagandist news articles use specific techniques to convey their message, such as whataboutism, red Herring, and name calling, among many others (cf. Section SECREF3). Whereas proving intent is not easy, we can analyse the language of a claim/article and look for the use of specific propaganda techniques. Going at this fine-grained level can yield more reliable systems and it also makes it possible to explain to the user why an article was judged as propagandist by an automatic system.
With this in mind, we organised the shared task on fine-grained propaganda detection at the NLP4IF@EMNLP-IJCNLP 2019 workshop. The task is based on a corpus of news articles annotated with an inventory of 18 propagandist techniques at the fragment level. We hope that the corpus would raise interest outside of the community of researchers studying propaganda. For example, the techniques related to fallacies and the ones relying on emotions might provide a novel setting for researchers interested in Argumentation and Sentiment Analysis.
The task is based on a corpus of news articles annotated with an inventory of 18 propagandist techniques at the fragment level.
|
What corpus does the task base on?
|
A corpus of news articles annotated with an inventory of 18 propagandist techniques at the fragment level.
|
null | false
| 264
|
Recently there has been a considerable interest in language modeling due to various academic and commercial demands. Academically, many studies have investigated this domain such as machine translation, chat-bot, message generation, image tagging and other language-related areas. Commercially, it can be used as a core technology for providing a new application on consumer products or services. For instance, an automatic message-reply prediction service can be launched in mobile devices, thus helping a user to send a reply message when he/she is not provided with a proper input interface.
To model the language of human dialogue, a recurrent neural network (RNN) structure is known to show the state of the arts performance with its ability to learn a sequential pattern of the data BIBREF0 . Among the RNN structures, a Long Short-Term Memory RNN (LSTM-RNN) and its variants are successfully used for language modeling tasks BIBREF1 , BIBREF2 . However, as a kind of deep learning technique, the LSTM-RNN and the RNN structure requires both a large number of data and huge computing power to train the model properly. Hence any attempts for applying the RNN structure to personalized language modeling are mainly constrained by the following two limitations. First, personal mobile devices contain private message data among close acquaintances, so users seldom agree to transfer their log out of the devices. This causes a limitation of gathering the whole user data to common computing spaces, where high-performance machines are available. Second, in relatively small computing machines, i.e., smart phone, it is not always-guaranteed to have enough resources to train a deep model within the devices.
To resolve these limitations, we propose fast transfer learning schemes. It trains a base model with a large dataset and copies its first n-many layers to the first n-many layers of a target model. Then the target model is fine-tuned with relatively small target data. Several learning schemes such as freezing a certain layer or adding a surplus layer are proposed for achieving the result. In experiments, we trained a general language model with huge corpus such as an Workshop on Statistical Machine Translation (WMT) data and a movie script data by using powerful computing machines, and then transferred the model to target environment for updating to be a personalized language model. With this approach, the final model can mimic target user's language style with proper syntax.
In the experiments, we trained the general language model with literary-style data and applied the transfer learning with spoken-style data. Then we evaluated the model output for sentence completion task in a qualitative and a quantitative manner. The test result showed that the model learned the style of the target language properly. Another test was conducted by training the general language model with the script of the drama, “Friends," and by applying transfer learning with main character corpora from the script to generate the personalized language model. The message-reply prediction task was evaluated with this model. The test result shows higher similarity between the output of the personalized language model and the same user dialogue than the one between the output of the personalized language model and other users' dialogues.
The contributions of this paper are as follows. First, we propose efficient transfer learning schemes for personalized language modeling, which is the first research on transfer learning for RNN based language models with privacy preserving. Second, we show the applicability of our research to the target scenario in the short message reply application by training the model in the similar environment to that of the mobile device, and highlight its test results.
First, we propose efficient transfer learning schemes for personalized language modeling, which is the first research on transfer learning for RNN based language models with privacy preserving.
|
Are the schemes proposed by them the first research on transfer learning for RNN based language models with privacy preserving?
|
Yes.
|
1911.03154
| false
| null |
In this paper, we study the problem of how to do simultaneous translation better with a pretrained vanilla CNMT model. We formulate simultaneous translation as two nested loops: an outer loop that updates input buffer with newly observed source tokens and an inner loop that translates source tokens in the buffer updated at each outer step. For the outer loop, the input buffer can be updated by an ASR system with an arbitrary update schedule. For the inner loop, we perform prefix translation using the pretrained CNMT model with dynamically built encoder and decoder hidden states. We also design two novel stopping criteria for the inner loop: Length and EOS (LE) controller that stops with heuristics, and Trainable (TN) controller that learns to stop with a better quality and latency balance. We evaluate our method on IWSLT16 German-English (DE-EN) translation in both directions, WMT15 English-German (EN-DE) translation in both directions, and NIST Chinese-to-English (ZH$\rightarrow $EN) translation. The result shows our method consistently improves over the de-facto baselines, and achieves low latency and reasonable BLEU scores.
We evaluate our method on IWSLT16 German-English (DE-EN) translation in both directions, WMT15 English-German (EN-DE) translation in both directions, and NIST Chinese-to-English (ZH$\rightarrow $EN) translation.
|
What corpora is used?
|
The answers are shown as follows:
* IWSLT16
* WMT15
* NIST
|
null | false
| null |
Sinigang is a Filipino soup or stew characterized by its sour and savory taste. It is most often associated with tamarind (Filipino: "sampalok"), although it can use other sour fruits and leaves as the souring agent
|
What is Sinigang?
|
Sinigang is a popular dish in the Philippines characterized by its sour, tangy, and savory taste. Tamarind is the most common souring agent used, but other sour fruits may be used as well. Various types of protein can be used in this soup dish, ranging from pork belly, shrimp, fish, and beef.
|
null | false
| null |
What kind of cars are good for families with 3 small children?
|
When shopping for a car that will accommodate three small children, you need to think about car seats and storage space. Each kid needs to have their own properly installed car seat and you’ll likely need to have enough room for strollers, diaper bags, and other gear. Some mid-size sedans and small SUVs can fit 3 car seats across the back row, but to fit them safely, you’ll need to find the right set of thin car seats the meet the size requirements of your children (e.g. height and weight). For example, Clek makes narrow car seats. That said, taking a trip with 3 kids in these mid-size cars will be challenging unless you add a luggage rack to the top of your car.
Mid-size SUVs like the Honda Pilot, Kia Telluride, and the Toyota Highlander can accommodate wider car seats in their second row and even include a third row. That said, the third row is difficult to access when car seats are installed in the second row and there isn’t much luggage space when the third row is used for seating.
All that considered, your best bet for 3 kids and lots of gear is a minivan or full size SUV (Chevy Suburban, GMC Yukon XL, Toyota Sequoia). Choosing between those is a matter of preference. It’s best to test drive the various options at local car dealerships.
|
|
null | false
| null |
Using Rene Girard's mimetic theory, which states that mutual desire for a coveted position or object among a group of individuals creates social rivalry and eventual scapegoating, write a short summary about a football team's season.
|
Season Start:
Player 1: I am the fastest. I should be the quarterback.
Player 2. I am the strongest. I should be the quarterback.
Player 3: I am the smartest. I should be the quarterback
Player 4. I have the most experience. I should be the quarterback.
Tensions grow among the players during tryouts as they vie for the quarterback position
Coach: We want the best players on the field in each position. We only have one quarterback. We need to work as a team to win the championship. We are going with Player 4 as the starter.
Season End:
The team wins most games but loses in the semi-final despite great effort and superb play from Player 4. However, the entire team is fed up with Player 4.
Coach to Player 4: We are going to move in a new direction next season. We are going with Player 2 and you are no longer on the team.
The team rejoices for their problems are "solved" with the removal of Player 4.
However, in actuality, the team is worse off than before. Scapegoating solves nothing.
|
|
null | false
| null |
In 2014, he starred in George Clooney's The Monuments Men, and played the minor role of scientist Dr. Mann, in Christopher Nolan's Interstellar. That same year, Damon appeared as a celebrity correspondent for Years of Living Dangerously In 2015, Damon portrayed the main character, astronaut Mark Watney, in Ridley Scott's The Martian, based on Andy Weir's best-selling novel of the same name, a role that earned him the Golden Globe Award for Best Actor – Motion Picture Musical or Comedy and his second Academy Award nomination for Best Actor. Having not returned for the fourth film in the Bourne film series, Damon reprised his role in 2016's Jason Bourne, reuniting with Paul Greengrass. In 2017, Damon played the lead role in Zhang Yimou's The Great Wall, a hit internationally and a disappointment at the domestic box office. The film, and Damon's casting, were not well received by critics. Later in 2017, he starred in two satires, George Clooney's 1950s-set Suburbicon, which was released in October, and Alexander Payne's comedy Downsizing, which was released in December In 2019, Damon portrayed Carroll Shelby in the action biographical drama Ford v Ferrari, directed by James Mangold.
|
From the passage provided, extract the movie that did well across the globe but received underwhelming traction in Matt Damon's home country.
|
Zhang Yimou's The Great Wall performed well globally but did not meet domestic box office expectations.
|
null | false
| null |
Roslyn is 36 years of age, weighs 74 kg, is 165 cm tall, and has 35% body fat. Her father had a heart attack at age 68 and her brother had an angioplasty at age 40. Both are living and well. Her blood chemistry values measure total cholesterol at 195 mg% and HDL cholesterol at 42 mg%. Her fasting blood sugar is 105 mg%. Roslyn is sedentary and has engaged in no endurance training program since college. Brainstorm an appropriate exercise training program to help Roslyn improve her C-R (cardiorespiratory) fitness. Be sure to calculate a target heart rate range with corresponding RPE.
|
We can prescribe an exercise program through the lens of FITT: frequency, intensity, time, and type. For frequency, we should aim for 3-5 days per week, with the higher the better the improvements for C-R fitness. As for intensity, we want to aim for 60-90% HRR (heart rate recovery), anywhere from 4-8 RPE (rate of perceived exertion). Since Roslyn is 36 years of age, her predicted maximum heart rate is 184bpm, and we are given that her RHR (resting heart rate) is 64bpm. Thus, we’re aiming for a target heart rate range of 136 BPM to 172 BPM. For time, the duration depends on intensity, but optimally, we want to be working out for at least 20 minutes and at most an hour. As for the type of exercise, we can assess the options that are available and feasible for Roslyn to perform, and choose between weight-bearing exercises (such as walking or jogging) and non-weight-bearing exercises such as swimming or rowing. Another thing to consider is that Roslyn is sedentary, and may need to slowly progress into a more intense workout. A sample exercise training program for Roslyn may first start with a moderate-intensity aerobic activity for 30 minutes a day, three times a week (Monday, Wednesday, and Friday). As she gets more acclimated to the exercise level, we want to progress to more vigorous exercise. According to David P. Swain, vigorous aerobic exercise may result in “substantially more benefits than moderate-intensity exercise”, which include but are not limited to greater improvements in aerobic fitness and greater reductions in coronary heart disease risk. Thus, we can add two more days of vigorous aerobic exercise for Roslyn, such as jogging for 20 minutes on Tuesdays and Thursdays. Another important aspect to consider is motivation, which we can help Roslyn maintain by switching up her forms of exercise to keep interest. For example, we can switch out jogging with cycling, or hiking hilly terrain.
|
|
null | false
| 200
|
Datasets and Models We evaluate our adversarial attacks on different text classification datasets from tasks such as sentiment classification, subjectivity detection and question type classification. Amazon, Yelp, IMDB are sentence-level sentiment classification datasets which have been used in recent work BIBREF15 while MR BIBREF16 contains movie reviews based on sentiment polarity. MPQA BIBREF17 is a dataset for opinion polarity detection, Subj BIBREF18 for classifying a sentence as subjective or objective and TREC BIBREF19 is a dataset for question type classification.
We use 3 popular text classification models: word-LSTM BIBREF20, word-CNN BIBREF21 and a fine-tuned BERT BIBREF12 base-uncased classifier. For each dataset we train the model on the training data and perform the adversarial attack on the test data. For complete model details refer to Appendix.
As a baseline, we consider TextFooler BIBREF11 which performs synonym replacement using a fixed word embedding space BIBREF22. We only consider the top $K{=}50$ synonyms from the MLM predictions and set a threshold of 0.8 for the cosine similarity between USE based embeddings of the adversarial and input text.
Results We perform the 4 modes of our attack and summarize the results in Table . Across datasets and models, our BAE attacks are almost always more effective than the baseline attack, achieving significant drops of 40-80% in test accuracies, with higher average semantic similarities as shown in parentheses. BAE-R+I is the strongest attack since it allows both replacement and insertion at the same token position, with just one exception. We observe a general trend that the BAE-R and BAE-I attacks often perform comparably, while the BAE-R/I and BAE-R+I attacks are much stronger. We observe that the BERT-based classifier is more robust to the BAE and TextFooler attacks than the word-LSTM and word-CNN models which can be attributed to its large size and pre-training on a large corpus.
The baseline attack is often stronger than the BAE-R and BAE-I attacks for the BERT based classifier. We attribute this to the shared parameter space between the BERT-MLM and the BERT classifier before fine-tuning. The predicted tokens from BERT-MLM may not drastically change the internal representations learned by the BERT classifier, hindering their ability to adversarially affect the classifier prediction.
Effectiveness We study the effectiveness of BAE on limiting the number of R/I operations permitted on the original text. We plot the attack performance as a function of maximum $\%$ perturbation (ratio of number of word replacements and insertions to the length of the original text) for the TREC dataset. From Figure , we clearly observe that the BAE attacks are consistently stronger than TextFooler. The classifier models are relatively robust to perturbations up to 20$\%$, while the effectiveness saturates at 40-50$\%$. Surprisingly, a 50$\%$ perturbation for the TREC dataset translates to replacing or inserting just 3-4 words, due to the short text lengths.
Qualitative Examples We present adversarial examples generated by the attacks on a sentence from the IMDB and Yelp datasets in Table . BAE produces more natural looking examples than TextFooler as tokens predicted by the BERT-MLM fit well in the sentence context. TextFooler tends to replace words with complex synonyms, which can be easily detected. Moreover, BAE's additional degree of freedom to insert tokens allows for a successful attack with fewer perturbations.
Human Evaluation We consider successful adversarial examples generated from the Amazon and IMDB datasets and verify their sentiment and grammatical correctness. Human evaluators annotated the sentiment and the grammar (Likert scale of 1-5) of randomly shuffled adversarial examples and original texts. From Table , BAE and TextFooler have inferior accuracies compared to the Original, showing they are not always perfect. However, BAE has much better grammar scores, suggesting more natural looking adversarial examples.
Ablation Study We analyze the benefits of R/I operations in BAE in Table . From the table, the splits $\mathbb {A}$ and $\mathbb {B}$ are the $\%$ of test points which compulsorily need I and R operations respectively for a successful attack. We can observe that the split $\mathbb {A}$ is larger than $\mathbb {B}$ thereby indicating the importance of the I operation over R. Test points in split require both R and I operations for a successful attack. Interestingly, split is largest for Subj, which is the most robust to attack (Table ) and hence needs both R/I operations. Thus, this study gives positive insights towards the importance of having the flexibility to both replace and insert words.
Refer to the Appendix for additional results, effectiveness graphs and details of human evaluation.
We use 3 popular text classification models: word-LSTM (Hochreiter and Schmidhuber, 1997), word-CNN (Kim, 2014) and a fine-tuned BERT (Devlin et al., 2018) base-uncased classifier.
|
What text classification models are used?
|
The authors use 3 popular text classification models: word-LSTM, word-CNN and a fine-tuned BERT base-uncased classifier.
|
null | false
| null |
What is "diglossia"?
|
Diglossia is when one community uses two different languages or dialects depending on the circumstances e.g. for written things and for casual conversation. A weird (not actual schoolbook) case of that is Swiss German. High German is used for written and formal things while Swiss German is used to speak.
|
|
null | false
| null |
Write a whimsical and funny horoscope reading that makes a baseball reference
|
The gods of justice say they will hear your petition just as soon as they have finished discussing the designated-hitter rule
|
|
null | false
| 309
|
Deep neural network (DNN) has aroused dramatically attention due to their extraordinary performance in computer vision BIBREF6 , BIBREF7 , speech recognition BIBREF8 and natural language processing (NLP) BIBREF9 tasks. We apply DNN to our Zhihu Live quality evaluation problem aiming to approximate a function INLINEFORM0 which can accurately predict a Zhihu Live's score.
In our quality evaluation task, we take multiple layer perception BIBREF8 as the basic composition block of MTNet. Since we treat the Zhihu Live quality evaluation problem as a regression task, we set the output neuron equal to 1. DNNs are trained by backpropagation algorithm BIBREF8 .
The calculation details of neural network can be illustrated as: DISPLAYFORM0
where INLINEFORM0 represents output of a neuron, INLINEFORM1 represents weights of the connections, INLINEFORM2 represents bias, INLINEFORM3 represents nonlinear activation function (sigmoid, tanh and ReLU are often used in practice).
Deep neural network (DNN) has aroused dramatically attention due to their extraordinary performance in computer vision [6, 13], speech recognition [4] and natural language processing (NLP) [15] tasks. We apply DNN to our Zhihu Live quality evaluation problem aiming to approximate a function f ∗ which can accurately predict a Zhihu Live’s score.
|
What network do they use? CNN OR RNN?
|
None of them. They use DNN.
|
null | false
| null |
Jim Robson OBC (born January 17, 1935) is a former radio and television broadcaster who was the play-by-play announcer of the Vancouver Canucks' games from 1970 to 1999.
|
Extract all the years from the following text.
|
1935, 1970, 1999
|
null | false
| 247
|
We compare with the following baseline methods:
RB (Rule based method): The rule based method proposed in BIBREF33 .
CB (Common-sense based method): This is the knowledge based method proposed by BIBREF34 . We use the Chinese Emotion Cognition Lexicon BIBREF35 as the common-sense knowledge base. The lexicon contains more than 5,000 kinds of emotion stimulation and their corresponding reflection words.
RB+CB+ML (Machine learning method trained from rule-based features and facts from a common-sense knowledge base): This methods was previously proposed for emotion cause classification in BIBREF36 . It takes rules and facts in a knowledge base as features for classifier training. We train a SVM using features extracted from the rules defined in BIBREF33 and the Chinese Emotion Cognition Lexicon BIBREF35 .
SVM: This is a SVM classifier using the unigram, bigram and trigram features. It is a baseline previously used in BIBREF24 , BIBREF31
Word2vec: This is a SVM classifier using word representations learned by Word2vec BIBREF32 as features.
Multi-kernel: This is the state-of-the-art method using the multi-kernel method BIBREF31 to identify the emotion cause. We use the best performance reported in their paper.
CNN: The convolutional neural network for sentence classification BIBREF5 .
Memnet: The deep memory network described in Section SECREF3 . Word embeddings are pre-trained by skip-grams. The number of hops is set to 3.
ConvMS-Memnet: The convolutional multiple-slot deep memory network we proposed in Section SECREF13 . Word embeddings are pre-trained by skip-grams. The number of hops is 3 in our experiments.
Table 2 shows the evaluation results. The rule based RB gives fairly high precision but with low recall. CB, the common-sense based method, achieves the highest recall. Yet, its precision is the worst. RB+CB, the combination of RB and CB gives higher the F-measure But, the improvement of 1.27% is only marginal compared to RB.
For machine learning methods, RB+CB+ML uses both rules and common-sense knowledge as features to train a machine learning classifier. It achieves F-measure of 0.5597, outperforming RB+CB. Both SVM and word2vec are word feature based methods and they have similar performance. For word2vec, even though word representations are obtained from the SINA news raw corpus, it still performs worse than SVM trained using n-gram features only. The multi-kernel method BIBREF31 is the best performer among the baselines because it considers context information in a structured way. It models text by its syntactic tree and also considers an emotion lexicon. Their work shows that the structure information is important for the emotion cause extraction task.
Naively applying the original deep memory network or convolutional network for emotion cause extraction outperforms all the baselines except the convolutional multi-kernel method. However, using our proposed ConvMS-Memnet architecture, we manage to boost the performance by 11.54% in precision, 4.84% in recall and 8.24% in F-measure respectively when compared to Memnet. The improvement is very significant with INLINEFORM0 -value less than 0.01 in INLINEFORM1 -test. The ConvMS-Memnet also outperforms the previous best-performing method, multi-kernel, by 3.01% in F-measure. It shows that by effectively capturing context information, ConvMS-Memnet is able to identify the emotion cause better compared to other methods.
It shows that by effectively capturing context information, ConvMS-Memnet is able to identify the emotion cause better compared to other methods.
|
Which model is able to identify the emotion cause better compared to other methods?
|
ConvMS-Memnet.
|
null | false
| null |
"Coffee is a beverage prepared from roasted coffee beans. Darkly colored, bitter, and slightly acidic, coffee has a stimulating effect on humans, primarily due to its caffeine content. It has the highest sales in the world market for hot drinks.
Seeds of the Coffee plant's fruits are separated to produce un-roasted green coffee beans. The beans are roasted and then ground into fine particles that are typically steeped in hot water before being filtered out, producing a cup of coffee. It is usually served hot, although chilled or iced coffee is common. Coffee can be prepared and presented in a variety of ways (e.g., espresso, French press, caffè latte, or already-brewed canned coffee). Sugar, sugar substitutes, milk, and cream are often used to mask the bitter taste or enhance the flavor."
|
What type of effect does Coffee have on humans?
|
Coffee has a stimulating effect on humans, primarily due to its caffeine content.
|
null | false
| null |
Which are the states in India with coastlines?
|
Most of the southern states in India are having coastlines that include Karnataka, Kerala, Tamil Nadu, Andra Pradesh, and Telangana. Towards the east-west Bengal and Orissa have a long coastline and towards the west Maharashtra and Gujrat.
|
|
null | false
| null |
Landscape Arch is a natural arch in Arches National Park, Utah, United States. It is among the longest natural rock arches in the world. The arch is among many in the Devils Garden area in the north of the park. Landscape Arch was named by Frank Beckwith who explored the area in the winter of 1933–1934 as the leader of an Arches National Monument scientific expedition. The arch can be reached by a 0.8 mi (1.3 km) graded gravel trail.
|
When was the Landscape Arch named and by whom?
|
Landscape Arch was named in 1933 by Frank Bekwith who explored it in the winter of 1933-1934 as the leader of an Arches National Monument scientific expedition.
|
null | false
| null |
Lawrence Grant White (September 26, 1887 – September 8, 1956) was an American architect, a partner in the architectural firm of McKim, Mead & White, co-founded by his father Stanford White, and for five years the president of the National Academy of Design.
|
Who was Lawrence Grant White?
|
Lawrence Grant White was born on September 26, 1887. He was an American Architect, and was the President of the National Academy of Design for 5 years. Larry, as known by his friends, was born in NY, and was the son of Sanford and Elizabeth White. Larry graduated from Harvard University in 1908. From 1915-1917 he worked at the firm McKim, Mead, and White.
|
null | false
| null |
The greater blue-ringed octopus (Hapalochlaena lunulata) is one of four species of extremely venomous blue-ringed octopuses belonging to the family Octopodidae. This particular species of blue-ringed octopus is known as one of the most toxic marine animals in the world.
|
Is the blue-ring octopus harmless?
|
No, it is extremely toxic.
|
null | false
| null |
Name a few great fantasy book series.
|
Any list of great fantasy series is sure to include the Lord of the Rings. Written in the mid 20th century by J.R.R Tolkien, the story centers on Frodo Baggins, a young hobbit tasked with the impossible task of carrying and ultimately destroying the One Ring, which was forged by Sauron as a means to claim complete dominion over all of Middle-earth. In his quest, Frodo is aided by three hobbits, a human ranger, a dwarf, an elf and a wizard. Together this band of friends and heroes battle orcs, trolls, wargs, giant spider and a host of dark creatures in their quest to save Middle-earth.
Another epic fantasy series is the Wheel of Time by Robert Jordan. Told over 14 volumes, this tale is set in a high fantasy world where magic exists, but only some can access it. The series begins with Moraine, a powerful channeler on a quest to find the prophesied "Dragon" who has been reborn with the power to defeat The Dark One. Moraine follows the clues to a small village and to five young men and women, one of whom she is convinced is The Dragon reborn. Throughout the series, these six, along with a coterie of interesting and intricately webbed characters quest throughout the world, fighting the emergence of dark forces and working to discover their own powers and destinies.
Arguably the most popular fantasy series ever written is Harry Potter by J.K. Rowling. The seven novels in this series follow the story of Harry Potter, a young orphan who discovers on his 11th birthday that he is a wizard. He is welcomed into the prestigious Hogwarts School of Witchcraft and Wizardry where he learns magics, makes wonderful friends and a few enemies. He also learns of and must faced his dark history and connection to Lord Voldemort, one of the most powerful, evil dark wizards who killed Harry's parents and tried to kill Harry when he was just an infant.
|
|
null | false
| null |
I have 3 apples, Jack has 4 pears and Jill has 5 books. After I ate 5 of Jack's pears, how many fruits we have in total now?
|
Jack has 4 pears in the beginning, which is less than 5. So you can't really ate 5 pears. The question itself is invalid and does not have an answer
|
|
null | false
| 74
|
Datasets. We perform our experiments with two predetermined event categories: cyber security (CyberAttack) and death of politicians (PoliticianDeath). These event categories are chosen as they are representative of important event types that are of interest to many governments and companies. The need to create our own dataset was motivated by the lack of public datasets for event detection on microposts. The few available datasets do not suit our requirements. For example, the publicly available Events-2012 Twitter dataset BIBREF20 contains generic event descriptions such as Politics, Sports, Culture etc. Our work targets more specific event categories BIBREF21. Following previous studies BIBREF1, we collect event-related microposts from Twitter using 11 and 8 seed events (see Section SECREF2) for CyberAttack and PoliticianDeath, respectively. Unlabeled microposts are collected by using the keyword `hack' for CyberAttack, while for PoliticianDeath, we use a set of keywords related to `politician' and `death' (such as `bureaucrat', `dead' etc.) For each dataset, we randomly select 500 tweets from the unlabeled subset and manually label them for evaluation. Table TABREF25 shows key statistics from our two datasets.
Comparison Methods. To demonstrate the generality of our approach on different event detection models, we consider Logistic Regression (LR) BIBREF1 and Multilayer Perceptron (MLP) BIBREF2 as the target models. As the goal of our experiments is to demonstrate the effectiveness of our approach as a new model training technique, we use these widely used models. Also, we note that in our case other neural network models with more complex network architectures for event detection, such as the bi-directional LSTM BIBREF17, turn out to be less effective than a simple feedforward network. For both LR and MLP, we evaluate our proposed human-AI loop approach for keyword discovery and expectation estimation by comparing against the weakly supervised learning method proposed by BIBREF1 (BIBREF1) and BIBREF17 (BIBREF17) where only one initial keyword is used with an expectation estimated by an individual expert.
Parameter Settings. We empirically set optimal parameters based on a held-out validation set that contains 20% of the test data. These include the hyperparamters of the target model, those of our proposed probabilistic model, and the parameters used for training the target model. We explore MLP with 1, 2 and 3 hidden layers and apply a grid search in 32, 64, 128, 256, 512 for the dimension of the embeddings and that of the hidden layers. For the coefficient of expectation regularization, we follow BIBREF6 (BIBREF6) and set it to $\lambda =10 \times $ #labeled examples. For model training, we use the Adam BIBREF22 optimization algorithm for both models.
Evaluation. Following BIBREF1 (BIBREF1) and BIBREF3 (BIBREF3), we use accuracy and area under the precision-recall curve (AUC) metrics to measure the performance of our proposed approach. We note that due to the imbalance in our datasets (20% positive microposts in CyberAttack and 27% in PoliticianDeath), accuracy is dominated by negative examples; AUC, in comparison, better characterizes the discriminative power of the model.
Crowdsourcing. We chose Level 3 workers on the Figure-Eight crowdsourcing platform for our experiments. The inter-annotator agreement in micropost classification is taken into account through the EM algorithm. For keyword discovery, we filter keywords based on the frequency of the keyword being selected by the crowd. In terms of cost-effectiveness, our approach is motivated from the fact that crowdsourced data annotation can be expensive, and is thus designed with minimal crowd involvement. For each iteration, we selected 50 tweets for keyword discovery and 50 tweets for micropost classification per keyword. For a dataset with 80k tweets (e.g., CyberAttack), our approach only requires to manually inspect 800 tweets (for 8 keywords), which is only 1% of the entire dataset.
Following Ritter et al. (2015) and Konovalov et al. (2017), we use accuracy and area under the precisionrecall curve (AUC) metrics to measure the performance of our proposed approach.
|
What metrics are used to measure the performance of their proposed approach?
|
AUC metrics.
|
null | false
| null |
Where was city musician Thomas Russell born?
|
Los Angeles
|
|
null | false
| 146
|
We pose the prediction task as a binary classification problem. Specifically, given an image and associated question, a system outputs a binary label indicating whether a crowd will agree on the same answer. Our goal is to design a system that can detect which visual questions to assign a disagreement label, regardless of the disagreement cause (e.g., subjectivity, ambiguity, difficulty). We implement both random forest and deep learning classifiers.
A visual question is assigned either an answer agreement or disagreement label. To assign labels, we employ 10 crowdsourced answers for each visual question. A visual question is assigned an answer agreement label when there is an exact string match for 9 of the 10 crowdsourced answers (after answer pre-preprocessing, as discussed in the previous section) and an answer disagreement label otherwise. Our rationale is to permit the possibility of up to one “careless/spam" answer per visual question. The outcome of our labeling scheme is that a disagreement label is agnostic to the specific cause of disagreement and rather represents the many causes (described above).
For our first system, we use domain knowledge to guide the learning process. We compile a set of features that we hypothesize inform whether a crowd will arrive at an undisputed, single answer. Then we apply a machine learning tool to reveal the significance of each feature. We propose features based on the observation that answer agreement often arises when 1) a lay person's attention can be easily concentrated to a single, undisputed region in an image and 2) a lay person would find the requested task easy to address.
We employ five image-based features coming from the salient object subitizing BIBREF22 (SOS) method, which produces five probabilities that indicate whether an image contains 0, 1, 2, 3, or 4+ salient objects. Intuitively, the number of salient objects shows how many regions in an image are competing for an observer's attention, and so may correlate with the ease in identifying a region of interest. Moreover, we hypothesize this feature will capture our observation from the previous study that counting problems typically leads to disagreement for images showing many objects, and agreement otherwise.
We employ a 2,492-dimensional feature vector to represent the question-based features. One feature is the number of words in the question. Intuitively, a longer question offers more information and we hypothesize additional information makes a question more precise. The remaining features come from two one-hot vectors describing each of the first two words in the question. Each one-hot vector is created using the learned vocabularies that define all possible words at the first and second word location of a question respectively (using training data, as described in the next section). Intuitively, early words in a question inform the type of answers that might be possible and, in turn, possible reasons/frequency for answer disagreement. For example, we expect “why is" to regularly elicit many opinions and so disagreement. This intuition about the beginning words of a question is also supported by our analysis in the previous section which shows that different answer types yield different biases of eliciting answer agreement versus disagreement.
We leverage a random forest classification model BIBREF23 to predict an answer (dis)agreement label for a given visual question. This model consists of an ensemble of decision tree classifiers. We train the system to learn the unique weighted combinations of the aforementioned 2,497 features that each decision tree applies to make a prediction. At test time, given a novel visual question, the trained system converts a 2,497 feature descriptor of the visual question into a final prediction that reflects the majority vote prediction from the ensemble of decision trees. The system returns the final prediction along with a probability indicating the system's confidence in that prediction. We employ the Matlab implementation of random forests, using 25 trees and the default parameters.
We next adapt a VQA deep learning architecture BIBREF24 to learn the predictive combination of visual and textual features. The question is encoded with a 1024-dimensional LSTM model that takes in a one-hot descriptor of each word in the question. The image is described with the 4096-dimensional output from the last fully connected layer of the Convolutional Neural Network (CNN), VGG16 BIBREF25 . The system performs an element-wise multiplication of the image and question features, after linearly transforming the image descriptor to 1024 dimensions. The final layer of the architecture is a softmax layer.
We train the system to predict (dis)agreement labels with training examples, where each example includes an image and question. At test time, given a novel visual question, the system outputs an unnormalized log probability indicating its belief in both the agreement and disagreement label. For our system's prediction, we convert the belief in the disagreement label into a normalized probability. Consequently, predicted values range from 0 to 1 with lower values reflecting greater likelihood for crowd agreement.
We pose the prediction task as a binary classification problem. Specifically, given an image and associated question, a system outputs a binary label indicating whether a crowd will agree on the same answer. Our goal is to design a system that can detect which visual questions to assign a disagreement label, regardless of the disagreement cause (e.g., subjectivity, ambiguity, difficulty). We implement both random forest and deep learning classifiers.
|
What classifiers are used when posing the prediction task as a binary classification problem?
|
Both random forest and deep learning classifiers.
|
null | false
| null |
Explain the answer to this riddle: What has four legs in the morning, two legs in the afternoon, and three legs in the evening?
|
This riddle is referring to a human. A human in infancy (morning) crawls using four limbs. A human adult (afternoon) walks on two legs. An elderly person (evening) walks on two legs and a cane.
|
|
null | false
| 176
|
The pre-trained language model, BERT BIBREF0 has led to a big breakthrough in various kinds of natural language understanding tasks. Ideally, people can start from a pre-trained BERT checkpoint and fine-tune it on a specific downstream task. However, the original BERT models are memory-exhaustive and latency-prohibitive to be served in embedded devices or CPU-based online environments. As the memory and latency constraints vary in different scenarios, the pre-trained BERT model should be adaptive to different requirements with accuracy retained to the largest extent. Existing BERT-oriented model compression solutions largely depend on knowledge distillation BIBREF1, which is inefficient and resource-consuming because a large training corpus is required to learn the behaviors of a teacher. For example, DistilBERT BIBREF2 is re-trained on the same corpus as pre-training a vanilla BERT from scratch; and TinyBERT BIBREF3 utilizes expensive data augmentation to fit the distillation target. The costs of these model compression methods are as large as pre-training and unaffordable for low-resource settings. Therefore, it is straight-forward to ask, can we design a lightweight method to generate adaptive models with comparable accuracy using significantly less time and resource consumption? In this paper, we propose LadaBERT (Lightweight adaptation of BERT through hybrid model compression) to tackle the raised questions. Specifically, LadaBERT is based on an iterative hybrid model compression framework consisting of weighting pruning, matrix factorization and knowledge distillation. Initially, the architecture and weights of student model are inherited from the BERT teacher. In each iteration, the student model is first compressed by a small ratio based on weight pruning and matrix factorization, and is then fine-tuned under the guidance of teacher model through knowledge distillation. Because weight pruning and matrix factorization help to generate better initial and intermediate status
in the knowledge distillation iterations, the accuracy and efficiency of model compression can be greatly improved.
We conduct extensive experiments on five public datasets of natural language understanding. As an example, the performance comparison of LadaBERT and state-of-the-art models on MNLI-m dataset is illustrated in Figure FIGREF1. We can see that LadaBERT outperforms other BERT-oriented model compression baselines at various model compression ratios. Especially, LadaBERT-1 outperforms BERT-PKD significantly under $2.5\times $ compression ratio, and LadaBERT-3 outperforms TinyBERT under $7.5\times $ compression ratio while the training speed is accelerated by an order of magnitude.
The rest of this paper is organized as follows. First, we summarizes the related works of model compression and their applications to BERT in Section SECREF2. Then, the methodology of LadaBERT is introduced in Section SECREF3, and experimental results are presented in Section SECREF4. At last, we conclude this work and discuss future works in Section SECREF5.
Initially, the architecture and weights of student model are inherited from the BERT teacher.
|
What are the architecture and weights of the student model inherited from?
|
The BERT teacher.
|
null | false
| null |
What is K8s?
|
Kubernetes, also known as K8s, is an open-source system for automating deployment, scaling, and management of containerized applications.
|
|
null | false
| null |
What is geospatial data?
|
Geographic data and information, also known as geospatial data, is defined in the ISO/TC 211 series of standards as data and information having an implicit or explicit association with a location relative to Earth (a geographic location or geographic position). It is also called geospatial data and information,[citation needed] georeferenced data and information,[citation needed] as well as geodata and geoinformation.[citation needed]
Approximately 90% of government sourced data has a location component. Location information (known by the many names mentioned here) is stored in a geographic information system (GIS).
There are also many different types of geodata, including vector files, raster files, geographic databases, web files, and multi-temporal data.
Spatial data or spatial information is broader class of data whose geometry is relevant but it is not necessarily georeferenced, such as in computer-aided design (CAD), see geometric modeling.
|
|
null | false
| null |
Before Morning is a 1933 American pre-Code crime drama directed by Arthur Hoerl, and starring Leo Carrillo, Lora Baxter, and Taylor Holmes. The film was adapted for the screen by Arthur Hoerl, from the 1933 Broadway play of the same name by Edward and Edna Riley.
|
Based on this paragraph, who directed "Before Morning?"
|
Arthur Hoerl
|
null | false
| null |
The evolution of triathlon as a distinct sport is difficult to trace with precision. Many, including triathlon historian and author Scott Tinley, consider events in early twentieth century France to be the beginnings of triathlon, with many three element multisport events of differing composition appearing, all called by different names. The earliest record for an event was from 1901 in Joinville-le-Pont, Val-de-Marne: it called itself "Les Trois Sports" (The Three Sports). It was advertised as an event for "the sportsmen of the time" and consisted of a run, a bicycle and a canoe segment. By 19 June 1921, the event in Joinville-le-Pont had become more like a standard triathlon, with the canoe segment being replaced with a swim. According to the newspaper L’Auto, the race consisted of a 3 km (1.9 mi) run, a 12 km (7.5 mi) bike ride and the crossing of the river Marne, all staged consecutively and without a break. Throughout the 1920s other bike, run, and swim events appeared in different cities, such as the “Course des Trois Sports” in Marseille and "La Course des Débrouillards" in Poissy. These multisport events would continue to slowly spread and grow in popularity: by 1934 "Les Trois Sports" was being hosted in the city of La Rochelle, though it consisted of three distinct events, swimming a 200 m (660 ft) channel crossing, a 10 km (6 mi) bike competition around the harbour of La Rochelle and the parc Laleu, and a 1.2 km (0.75 mi) run in the stadium André-Barbeau. Throughout this growth with new events appearing no unified rules ever existed and as a whole triathlon would remain a minority event on the world stage
|
Given this paragraph about the history of triathlon, what is the earliest known triathlon event.
|
Triathlon events have evolved of the years, and its origins are not fully known. Triathlon historians have evidenced multisports events dating back to 1901. In this era "Les Trois Sports" did not include the swim as it does in modern day triathlon, but instead included a canoe discipline. As the name suggest, triathlon appears to have first surfaced in the history books in france in the suburbs of paris known as Joinville-le-pont, Val-de-marne.
|
null | false
| 231
|
We consider three datasets, two of which are a contribution of this work.
Dataset I consists of 480 concepts previously analyzed in BIBREF1, BIBREF4. 240 are positive examples, titles from the Wikipedia list of controversial issues, and 240 are negative examples chosen at random and exclusive of the positives. Over this dataset, we compare the methodology suggested here to those reported by BIBREF1, BIBREF4. As the latter report overall accuracy of their binary prediction, we convert our controversiality estimates to a binary classification by classifying the higher-scored half as controversial, and the lower half as non-controversial.
Dataset II is based on a more recent version of the Wikipedia list of controversial issues (May 2017). As positive examples we take, from this list, all concepts which appear more than 50 times in Wikipedia. This leaves 608 controversial Wikipedia concepts. For negative examples, we follow BIBREF1, BIBREF4 and select a like number of concepts at random. Here too, since each concept only has a binary label, we convert our estimation into a binary classification, and report accuracy.
Dataset III is extracted from 3561 concepts whose Wikipedia pages are under edit protection, assuming that many of them are likely to be controversial. They were then crowd-annotated, with 10 or more annotators per concept. The annotation instructions were: “Given a topic and its description on Wikipedia, mark if this is a topic that people are likely to argue about.”. Average pairwise kappa agreement on this task was 0.532. Annotations were normalized to controversiality scores on an integer scale of 0 - 10. We used this dataset for testing the models trained on Dataset I.
In all datasets, to obtain the sentence-level context of the concepts (positive and negative), we randomly select two equal-sized sets of Wikipedia sentences, that explicitly reference these concepts – i.e., that contain a hyperlink to the article titled by the concept. Importantly, in each sentence we mask the words that reference the concept – i.e., the surface form of the hyperlink leading to the concept – by a fixed, singular token, thus focusing solely on the context within which the concepts are mentioned.
We consider three datasets, two of which are a contribution of this work.
|
How many datasets do they consider?
|
Three datasets
|
null | false
| null |
Primary Insurance Amount and Monthly Benefit Amount calculations
Main article: Primary Insurance Amount
Workers in Social Security covered employment pay FICA (Federal Insurance Contributions Act) or SECA (Self Employed Contributions Act) taxes and earn quarters of coverage if earnings are above minimum amounts specified in the law. Workers with 40 quarters of coverage (QC) are "fully insured" and eligible for retirement benefits. Retirement benefit amounts depend upon the average of the person's highest 35 years of "adjusted" or "indexed" earnings. A person's payroll-taxable earnings from earlier years are adjusted for economy-wide wage growth, using the national average wage index (AWI), and then averaged. If the worker has fewer than 35 years of covered earnings these non-contributory years are assigned zero earnings. The sum of the highest 35 years of adjusted or indexed earnings divided by 420 (35 years times 12 months per year) produces a person's Average Indexed Monthly Earnings or AIME.
The AIME is then used to calculate the Primary Insurance Amount or PIA. For workers who turn 62 in 2021, the PIA computation formula is:
(a) 90 percent of the first $996 of average indexed monthly earnings, plus
(b) 32 percent of average indexed monthly earnings between $996 and $6,002, plus
(c) 15 percent of average indexed monthly earnings over $6,002
For workers who turn 62 in the future, the 90, 32, and 15 percent factors in the computation formula will remain the same but the dollar amounts in the formula (called bend points) will increase by wage growth in the national economy, as measured by the AWI. Because the AIME and the PIA calculation incorporate the AWI, Social Security benefits are said to be wage indexed. Because wages typically grow faster than prices, the PIAs for workers turning 62 in the future will tend to be higher in real terms but similar relative to average earnings in the economy at the time age 62 is attained.
Monthly benefit amounts are based on the PIA. Once the PIA is computed, it is indexed for price inflation over time. Thus, Social Security monthly benefit amounts retain their purchasing power throughout a person's retirement years.
A worker who first starts receiving a retirement benefit at the full retirement age receives a monthly benefit amount equal to 100 percent of the PIA. A worker who claims the retirement benefit before the full retirement age receives a reduced monthly benefit amount and a worker who claims at an age after the full retirement age (up to age 70) receives an increased monthly amount.
The 90, 32, and 15 percent factors in the PIA computation lead to higher replacement rates for persons with lower career earnings. For example, a retired individual whose average earnings are below the first bend point can receive a monthly benefit at the full retirement age that equals 90 percent of the person's average monthly earnings before retirement. The table shows replacement rates for workers who turned 62 in 2013.
The PIA computation formula for disabled workers parallels that for retired workers except the AIME is based on fewer years to reflect disablement before age 62. The monthly benefit amount of a disabled worker is 100 percent of PIA.
Benefits for spouses, children, and widow(er)s depend on the PIAs of a spouse or a deceased spouse. Aged spouse and divorced spouse beneficiaries can receive up to 50 percent of the PIA. Survivor benefit rates are higher and aged widow(er)s and aged surviving divorced spouses can receive 100 percent of the PIA.
Federal, state and local employees who have elected (when they could) NOT to pay FICA taxes are eligible for a reduced FICA benefits and full Medicare coverage if they have more than forty quarters of qualifying Social Security covered work. To minimize the Social Security payments to those who have not contributed to FICA for 35+ years and are eligible for federal, state and local benefits, which are usually more generous, Congress passed the Windfall Elimination Provision, WEP. The WEP provision will not eliminate all Social Security or Medicare eligibility if the worker has 40 quarters of qualifying income, but calculates the benefit payments by reducing the 90% multiplier in the first PIA bendpoint to 40–85% depending on the number of Years of Coverage. Foreign pensions are subject to WEP.
A special minimum benefit, based on an alternative PIA calculation, is available for some workers with long careers but low earnings. However, it is rarely higher than the regularly-computed PIA and thus few workers qualify for the special minimum benefit. Only 32,000 individuals received the special minimum benefit in 2019.
The benefits someone is eligible for are potentially so complicated that potential retirees should consult the Social Security Administration directly for advice. Many questions are addressed and at least partially answered on many online publications and online calculators.
|
Based on the reference text, provide a summary of how the monthly benefit amount for social security is calculated.
|
The monthly benefit amount for social security is calculated by determining an individual's highest 35 years of "indexed" earnings. The sum of the highest 35 years of "indexed" earnings is divided by 420, which is the number of months in 35 years, to determine an individuals Average Indexed Monthly Earnings, which is referred to as AIME. Once an individuals AIME is determined, the formula to calculate the monthly benefit is as follows:
- 90 percent of the first $996 of AIME, plus
- 32 percent of AIME between $996 and $6,002, plus
- 15 percent of AIME over $6,002
|
2004.04228
| false
| null |
For QA quality, we answer this question by training QA models of varying quality by fine-tuning different versions of BERT on SQuAD. We present results in Table . The QA models perform similarly despite substantially different performances on the SQuAD development set. Surprisingly, using the best QA model (bert-large-wwm) does not lead to the best correlations with human judgments. On CNN/DM, bert-large-wwm slightly underperforms bert-base and bert-large. On XSUM, bert-base slightly outperforms the other two BERT variants. These results indicate that QAGS is fairly robust to the quality of the underlying QA model, though we note that BERT is a strong QA baseline, and using weaker QA models might lead to larger performance dropoffs.
Surprisingly, using the best QA model (bert-large-wwm) does not lead to the best correlations with human judgments. On CNN/DM, bert-large-wwm slightly underperforms bert-base and bert-large.
|
What models are evaluated with QAGS?
|
The answers are shown as follows:
* bert-large-wwm
* bert-base
* bert-large
|
null | false
| null |
What are the words of House Swyft?
|
"Awake! Awake!"
|
|
null | false
| 251
|
As we showed that pre-trained language models are indeed very useful in clustering sentence representations by domains in an unsupervised manner, we now seek to harness this property for a down-stream task – domain data selection for machine translation. Domain data selection is the task of selecting examples from a large corpus which are as close as possible to the domain of interest, given a smaller set of in-domain examples. The selected examples can be used to either (1) train a domain-specific model from scratch BIBREF5, (2) fine-tune a pre-trained general-domain model BIBREF26, BIBREF7, or (3) prioritize data for annotation as in an Active-Learning framework, if only monolingual data is available BIBREF27. To demonstrate the need for domain data selection and set the stage for our data selection experiments, we perform preliminary experiments with NMT in a multi-domain scenario.
As we showed that pre-trained language models are indeed very useful in clustering sentence representations by domains in an unsupervised manner, we now seek to harness this property for a downstream task – domain data selection for machine translation.
|
Why pre-trained language models are indeed very useful in clustering sentences?
|
For it domains in an unsupervised manner.
|
null | false
| null |
Where is Tiger Woods from?
|
Cypress, California
|
|
null | false
| null |
Which is a species of fish? Pickerel or Pickle
|
Pickerel
|
|
null | false
| null |
Sabapathy is a 1941 Indian Tamil-language comedy film directed by A. V. Meiyappan and A. T. Krishnaswamy, and produced by Meiyappan. An adaptation of Pammal Sambandha Mudaliar's farce play of the same name, the film stars T. R. Ramachandran, Kali N. Rathnam, C. T. Rajakantham and K. Sarangapani. It focuses on the antics of two dim-witted men named Sabapathy: a wealthy man and his servant. The film was released on 14 December 1941 and became a commercial success.
|
Given this paragraph, tell me the storyline of Sabapathy.
|
Sabapathy is a comedy film about the antics of a wealthy man and his servant.
|
null | false
| null |
What movie won best picture at the Oscars in 2023
|
The movie Everything Everywhere All at Once won the Oscar for 2022 in the year 2023. In addition to winning best picture it was also the overall winner of the night, winning multiple awards including Best Directing, Best Actress in a Leading Role, Best Actor in a Supporting Role, Best Actress in a Supporting Role, etc. The movie was produced by A24 which is well known for producing excellent content
|
|
null | false
| null |
What is the best color ?
|
The best color should be black and white
|
|
null | false
| 14
|
We study textual autocomplete---the task of predicting a full sentence from a partial sentence---as a human-machine communication game. Specifically, we consider three competing goals for effective communication: use as few tokens as possible (efficiency), transmit sentences faithfully (accuracy), and be learnable to humans (interpretability). We propose an unsupervised approach which tackles all three desiderata by constraining the communication scheme to keywords extracted from a source sentence for interpretability and optimizing the efficiency-accuracy tradeoff. Our experiments show that this approach results in an autocomplete system that is 52% more accurate at a given efficiency level compared to baselines, is robust to user variations, and saves time by nearly 50% compared to typing full sentences.
Our experiments show that this approach results in an autocomplete system that is 52% more accurate at a given efficiency level compared to baselines, is robust to user variations, and saves time by nearly 50% compared to typing full sentences.
|
Is the approach robust to user variation？
|
Yes.
|
null | false
| null |
Who won the MLB World Series in 2020?
|
Los Angeles Dodgers
|
|
null | false
| null |
Is there a relation between wavelength and frequency?
|
Wavelength is inversely proportional to frequency. Wavelength is the distance between crests of waves whereas Frequency is number of waves occur in given time. Higher the frequency , lower the wavelength and vice-versa
|
|
null | false
| null |
Big Trouble in Little China (also known as John Carpenter's Big Trouble in Little China) is a 1986 American fantasy action-comedy film directed by John Carpenter, and starring Kurt Russell, Kim Cattrall, Dennis Dun and James Hong. The film tells the story of truck driver Jack Burton (Russell), who helps his friend Wang Chi (Dun) rescue Wang's green-eyed fiancée from bandits in San Francisco's Chinatown. They go into the mysterious underworld beneath Chinatown, where they face an ancient sorcerer named David Lo Pan (Hong), who requires a woman with green eyes to marry him in order to be released from a centuries-old curse.
|
What year was the movie Big Trouble in Little China released?
|
1986
|
null | false
| null |
An exoplanet or extrasolar planet is a planet outside the Solar System. The first possible evidence of an exoplanet was noted in 1917 but was not recognized as such. The first confirmation of detection occurred in 1992. A different planet, initially detected in 1988, was confirmed in 2003. As of 1 April 2023, there are 5,346 confirmed exoplanets in 3,943 planetary systems, with 855 systems having more than one planet. The James Webb Space Telescope (JWST) is expected to discover more exoplanets, and also much more about exoplanets, including composition, environmental conditions and potential for life.
|
In what year was the evidence of an exoplanet first confirmed?
|
The detection of an exoplanet was first confirmed in 1992
|
null | false
| null |
Can AI systems fall in love?
|
AI, unlike humans, has no emotions or feelings. It does, however, have the unique ability to perceive and analyze human sentiments from vast amounts of data on which it has been trained. It can even outperform humans when it comes to mimicking and reproducing these emotions in a more efficient and effective manner. While AI does not experience emotions in the same way that humans do, it is still capable of understanding and replicating them in a remarkable and unprecedented way.
|
|
null | false
| null |
What are some fun things to do with my 6 year old son
|
There are a number of interesting things you can do with your 6 year old son such as playing sports, card games, going to the park or visiting a museum
|
|
null | false
| null |
What's the appeal of a vacation rental on the Big Island of Hawaii?
|
The Big Island of Hawaii is a bit quieter than the island of Oahu, where Waikiki Beach is. While Waikiki is a beautiful beach and place, the Big Island of Hawaii has so much to offer that is unique only to this island. First, you can visit black sand beaches. This is due to all of the volcanic activity on the island. There are still active volcanoes on the island. You can also take tours around them should you wish to get closer. You still get all the beautiful beaches Hawaii offers, but with features, you can't find on the other islands. Also, if you want a slightly less touristy destination, the Big Island is for you!
|
|
null | false
| null |
Smelting is a process of applying heat to an ore, to extract a base metal. It is a form of extractive metallurgy. It is used to extract many metals from their ores, including silver, iron, copper, and other base metals. Smelting uses heat and a chemical- reducing agent to decompose the ore, driving off other elements as gases or slag and leaving the metal base behind. The reducing agent is commonly a fossil fuel source of carbon, such as coke—or, in earlier times, charcoal.The oxygen in the ore binds to carbon at high temperatures as the chemical potential energy of the bonds in carbon dioxide (CO2) is lower than the bonds in the ore.
|
Give me a one line summary about smelting
|
Smelting is the process of extracting metals from their ores by applying heat and a chemical-reducing agent to drive off non-wanted materials as gases and leaving the base (wanted) material behind.
|
null | false
| null |
Which Game of Thrones episode does Arya go blind?
|
Arya goes temporarily blind in the last episode of season 5, entitled "Mother's Mercy"
|
|
null | false
| 89
|
We utilize an off-the-shelf toolbox of OpenIE to the derive structured answer-relevant relations from sentences as to the point contexts. Relations extracted by OpenIE can be represented either in a triple format or in an n-ary format with several secondary arguments, and we employ the latter to keep the extractions as informative as possible and avoid extracting too many similar relations in different granularities from one sentence. We join all arguments in the extracted n-ary relation into a sequence as our to the point context. Figure FIGREF5 shows n-ary relations extracted from OpenIE. As we can see, OpenIE extracts multiple relations for complex sentences. Here we select the most informative relation according to three criteria in the order of descending importance: (1) having the maximal number of overlapped tokens between the answer and the relation; (2) being assigned the highest confidence score by OpenIE; (3) containing maximum non-stop words. As shown in Figure FIGREF5, our criteria can select answer-relevant relations (waved in Figure FIGREF5), which is especially useful for sentences with extraneous information. In rare cases, OpenIE cannot extract any relation, we treat the sentence itself as the to the point context.
Table TABREF8 shows some statistics to verify the intuition that the extracted relations can serve as more to the point context. We find that the tokens in relations are 61% more likely to be used in the target question than the tokens in sentences, and thus they are more to the point. On the other hand, on average the sentences contain one more question token than the relations (1.86 v.s. 2.87). Therefore, it is still necessary to take the original sentence into account to generate a more accurate question.
Here we select the most informative relation according to three criteria in the order of descending importance:
|
How many criteria are set for the selection of the most informative relation?
|
Three.
|
null | false
| null |
Why would people prefer cats over dogs?
|
Cats are easy to care for, quiet, and relatively affordable. They even groom themselves. There are certainly benefits to owning a cat that you won't get from owning a dog.
|
|
1811.02076
| false
| null |
We build on the state-of-the-art publicly available question answering system by docqa. The system extends BiDAF BIBREF4 with self-attention and performs well on document-level QA. We reuse all hyperparameters from docqa with the exception of number of paragraphs sampled in training: 8 instead of 4. Using more negative examples was important when learning from both fine and coarse annotations. The model uses character embeddings with dimension 50, pre-trained Glove embeddings, and hidden units for bi-directional GRU encoders with size 100. Adadelta is used for optimization for all methods. We tune two hyperparameters separately for each condition based on the held-out set: (1) $\alpha \in \lbrace .01, .1, .5, 1, 5, 10, 100 \rbrace $ , the weight of the coarse loss, and (2) the number of steps for early stopping. The training time for all methods using both coarse and fine supervision is comparable. We use Adadelta for optimization for all methods.
We build on the state-of-the-art publicly available question answering system by docqa. The system extends BiDAF BIBREF4 with self-attention and performs well on document-level QA.
|
What is the underlying question answering algorithm?
|
The answers are shown as follows:
* The system extends BiDAF BIBREF4 with self-attention
|
null | false
| null |
Voorhees Chapel is one of two chapels on the campus of Rutgers, The State University of New Jersey in New Brunswick, New Jersey. Built in 1925 with a donation from Elizabeth Rodman Voorhees, wife of Rutgers trustee Ralph Voorhees, the chapel once served the community of Douglass College. Douglass, founded the New Jersey College for Women (founded in 1918), was the women's residential college at Rutgers.
|
Which chapel on Rutger's campus was built 7 years after the New Jersey College for Women?
|
The Voorhees Chapel was built in 1925, seven years after the New Jersey College for Women was founded by Elizabeth Rodman Voorhees.
|
1906.10519
| false
| null |
We compare Blse (Sections UID23 – UID30 ) to VecMap, Muse, and Barista (Section "Previous Work" ) as baselines, which have similar data requirements and to machine translation (MT) and monolingual (Mono) upper bounds which request more resources. For all models (Mono, MT, VecMap, Muse, Barista), we take the average of the word embeddings in the source-language training examples and train a linear SVM. We report this instead of using the same feed-forward network as in Blse as it is the stronger upper bound. We choose the parameter $c$ on the target language development set and evaluate on the target language test set.
We compare Blse (Sections UID23 – UID30 ) to VecMap, Muse, and Barista (Section "Previous Work" ) as baselines, which have similar data requirements and to machine translation (MT) and monolingual (Mono) upper bounds which request more resources.
|
what baseline do they compare to?
|
The answers are shown as follows:
* VecMap
* Muse
* Barista
|
null | false
| null |
Asimov published three volumes of autobiography. In Memory Yet Green (1979) and In Joy Still Felt (1980) cover his life up to 1978. The third volume, I. Asimov: A Memoir (1994), covered his whole life (rather than following on from where the second volume left off). The epilogue was written by his widow Janet Asimov after his death. The book won a Hugo Award in 1995. Janet Asimov edited It's Been a Good Life (2002), a condensed version of his three autobiographies. He also published three volumes of retrospectives of his writing, Opus 100 (1969), Opus 200 (1979), and Opus 300 (1984).
|
In a section labeled "Isaac Asimov's Autobiography Volumes", from the given paragraph list the names of the volumes of Asimov's autobiography in chronological order with the following format {YEAR} - {TITLE} separated by newlines. Then, in a new section, list Janet Asimov's contributions in a section labeled "Asimov Biographies Janet Asimov Wrote or Contributed To" in the same format.
|
Isaac Asimov's Autobiography Volumes
1979 - In Memory Yet Green
1980 - In Joy Still Felt
1994 - I. Asimov: A Memoir
Asimov Biographies Janet Asimov Wrote or Contributed To
1994 - I. Asimov: A Memoir
2002 - It's Been a Good Life
|
null | false
| 514
|
State-of-the-art (SOTA) artificial neural networks (ANNs) achieve impressive results in a variety of machine intelligence tasks. However, they largely rely on mechanisms that diverge from the original inspiration from biological neural networks. As a result, only a small part of this prolific field also contributes to computational neuroscience. In fact, this biological implausibility is also an important issue for machine intelligence. For their impressive performance, ANNs trade off other desired properties, which are present in biological systems. For example, ANN training often demands very large and labelled datasets. When labels are unavailable, self-supervised learning schemes exist, where supervisory error signals generated by the network itself are exploited and backpropagated from the output towards the input to update the network's parameters. However, this global propagation of signals in deep networks introduces another limitation. Namely, it prevents the implementation of efficient distributed computing hardware that would be based on only local signals from neighbouring physical nodes in the network, and is in contrast to local synaptic plasticity rules that partly govern biological learning. Several pieces of work have been addressing parts of the biological implausibility and hardware-inefficiency of backpropagation in ANNs. such as the need for exactly symmetric forward and backward weights or the waiting time caused by the network's forward-backward pass between two training updates in a layer (weight transport and update-locking problems). Recently, an approximation to backpropagation that is mostly Hebbian, i.e. relies on mostly pre-and post-synaptic activity of each synapse, has been achieved by reducing the global error requirements to 1-bit information. Two schemes that further localize the signal that is required for a weight update are Equilibrium Propagation and Predictive Coding. Both methods approximate backpropagation through Hebbian-like learning, by delegating the global aspect of the computation, from a global error signal, to a global convergence of the network state to an equilibrium. This equilibrium is reached through several iterative steps of feed-forward and feed-back communication throughout the network, before the ultimate weight update by one training example. The biological plausibility and hardware-efficiency of this added iterative process of signal propagation are open questions that begin to be addressed.
Moreover, learning through backpropagation, and presumably also its approximations, has another indication of biological implausibility, which also significantly limits ANN applicability. Namely, it produces networks that are confused by small adversarial perturbations of the input, which are imperceptible by humans. It has recently been proposed that a defence strategy of "deflection" of adversarial attacks may be the ultimate solution to that problem. Through this strategy, to cause confusion in the network's inferred class, the adversary is forced to generate such a changed input that really belongs to the distribution of a different input class. Intuitively, but also strictly by definition, this deflection is achieved if a human assigns to the perturbed input the same label that the network does. Deflection of adversarial attacks in ANNs has been demonstrated by an elaborate scheme that is based on detecting the attacks. However, the human ability to deflect adversarial perturbations likely does not rely on detecting them, but rather on effectively ignoring them, making the deflecting type of robustness an emergent property of biological computation rather than a defence mechanism. The biological principles that underlie this property of robustness are unclear, but it might emerge from the distinct algorithms that govern learning in the brain.
Therefore, what is missing is a biologically plausible model that can learn from fewer data-points, without labels, through local plasticity, and without feedback from distant layers. This model could then be tested for emergent adversarial robustness. A good candidate category of biological networks and learning algorithms is that of competitive learning. Neurons that compete for their activation through lateral inhibition are a common connectivity pattern in the superficial layers of the cerebral cortex). This pattern is described as winner-take-all (WTA), because competition suppresses activity of weakly activated neurons, and emphasizes strong ones. Combined with Hebbian-like plasticity rules, WTA connectivity gives rise to competitivelearning algorithms. These networks and learning schemes have been long studied (Von der and a large literature based on simulations and analyses describes their functional properties. A WTA neuronal layer, depending on its specifics, can restore missing input signals, perform decision making i.e. winner selection, and generate oscillations such as those that underlie brain rhythms. Perhaps more importantly, its neurons can learn to become selective to different input patterns, such as orientation of visual bars in models of the primary visual cortex (Von der, MNIST handwritten digits, CIFAR10 objects, spatiotemporal spiking patterns, and can adapt dynamically to model changing objects. The WTA model is indeed biologically plausible, Hebbian plasticity is local, and learning is input-driven, relying on only feed-forward communication of neurons -properties that seem to address several of the limitations of ANNs. However, the model's applicability is limited to simple tasks. That is partly because the related theoretical literature remains surprisingly unsettled, despite its long history, and the strong and productive community interest. described a very related theory but for a model that is largely incompatible with ANNs and thus less practical. It uses spiking and stochastic neurons, input has to be discretized, and each input feature must be encoded through multiple binary neurons. Moreover, it was only proven for neurons with an exponential activation function. It remains therefore unclear which specific plasticity rule and structure could optimize an ANN WTA for Bayesian inference. It is also unclear how to minimize a common loss function such as cross-entropy despite unsupervised learning, and how a WTA could represent varying families of probability distributions. In summary, on the theoretical side, an algorithm that is simultaneously normative, based on WTA networks and Hebbian unsupervised plasticity, performs Bayesian inference, and, importantly, is composed of conventional, i.e. non-spiking, ANN elements and is rigorously linked to modern ANN tools such as cross-entropy loss, would be an important advance but has been missing. On the practical side, evidence that Hebbian WTA networks could be useful for presently pertinent issues of modern ANNs such as adversarial robustness, generation of synthetic images, or faster learning, has remained limited. Here we aim to fill these gaps. Recently, when WTA networks were studied in a theoretical framework compatible with conventional machine learning (ML), but in the context of short-term as opposed to long-term Hebbian plasticity, it resulted in surprising practical advantages over supervised ANNs. A similar theoretical approach could also reveal unknown advantages of long-term Hebbian plasticity in WTA networks. In addition, it could provide insights into how a WTA microcircuit could participate in larger-scale computation by deeper cortical or artificial networks.
Here we construct "SoftHebb", a biologically plausible WTA model that is based on standard ratebased neurons as in ANNs, can accommodate various activation functions, and learns without labels, using local plasticity and only feed-forward communication, i.e. the properties we seek in an ANN. Importantly, it is equipped with a simple normalization of the layer's activations, and an optional temperature-scaling mechanism, producing a soft WTA instead of selecting a single "hard" winner neuron. This allows us to prove formally that a SoftHebb layer is a generative mixture model that objectively minimizes its Kullback-Leibler (KL) divergence from the input distribution through Bayesian inference, thus providing a new formal ML-theoretic perspective of these networks. We complement our main results, which are theoretical, with experiments that are small-scale but produce intriguing results. As a generative model, SoftHebb has a broader scope than classification, but we test it on image classification tasks. Surprisingly, in addition to overcoming several inefficiencies of backpropagation, the unsupervised WTA model also outperforms a supervised two-layer perceptron in several aspects: learning speed and accuracy in the first presentation of the training dataset, robustness to noisy data and to one of the strongest white-box adversarial attacks, i.e. projected gradient descent (PGD), and without any explicit defence. Interestingly, the SoftHebb model also exhibits inherent properties of deflection of the adversarial attacks, and generates object interpolations.
Here we construct "SoftHebb", a biologically plausible WTA model that is based on standard ratebased neurons as in ANNs, can accommodate various activation functions, and learns without labels, using local plasticity and only feed-forward communication, i.e. the properties we seek in an ANN.****Some of our main contributions are theoretical and their proof is fully reproducible by following the rigorous derivations in the main text and in the appendices.
|
The equivalency between mixture model and WTA has been proposed in a previous work (Moraitis et al. 2020) , is the novelty on that end quite limited?
|
The novel theoretical result is not that. Our main results are (a) that such an ANN/probabilistic model can be optimized by a Hebbian plasticity rule, and (b) the derivation of the specific rule.** This is new, it was a key missing step for the normative design of larger future Hebbian learning models, the derivation is not trivial at all, and this theory already led to the study and demonstration of novel practical advantages of speed and robustness.
|
null | false
| null |
Who is the greatest Russian writer?
|
Alexander Sergeyevich Pushkin was a Russian author of the Romantic era who is considered by many to be the greatest Russian poet and the founder of modern Russian literature.
|
|
null | false
| null |
How many miles per year does the average person drive?
|
13,500
|
|
null | false
| null |
Who wrote Mahabharata?
|
According to ancient Indian tradition, Vyasa is the author of Mahabharata who is also a major character in the epic. It is first recited the sage Vaisampayana to the King Janamejaya who was the great-grandson of the Pandavas. The story is then recited again by a Ugrasrava Sauti to an assembly of sages in Naimisa forest.
|
|
null | false
| null |
What rock icon was the founder of The Society for the Prevention of Cruelty to Long-haired Men?
|
David Bowie.
|
|
null | false
| null |
Why is recycling important
|
The planet has a finite amount of resources and it is important to use these wisely to ensure availability for future generations. Recycling also reduces the amount of waste which will otherwise end up in landfills or be burnt for energy recovery.
|
|
null | false
| null |
How does a human do arithmetic calculation as fast as a calculator?
|
Use a tech called Mental abacus. When a person reads the numbers, he or she will mentally visualize an abacus and do the calculation by moving the beads in the abacus. The calculation can be done in a great speed.
|
|
1706.07179
| false
| null |
We evaluate the model's performance on the bAbI tasks BIBREF18 , a collection of 20 question answering tasks which have become a benchmark for evaluating memory-augmented neural networks. We compare the performance with the Recurrent Entity Networks model (EntNet) BIBREF17 . Performance is measured in terms of mean percentage error on the tasks.
The results are shown in Table 1 . The RelNet model achieves a mean error of 0.285% across tasks which is better than the results of the EntNet model BIBREF17 . The RelNet model is able to achieve 0% test error on 11 of the tasks, whereas the EntNet model achieves 0% error on 7 of the tasks.
We compare the performance with the Recurrent Entity Networks model (EntNet) BIBREF17 .
The RelNet model achieves a mean error of 0.285% across tasks which is better than the results of the EntNet model BIBREF17 . The RelNet model is able to achieve 0% test error on 11 of the tasks, whereas the EntNet model achieves 0% error on 7 of the tasks.
|
What methods is RelNet compared to?
|
The answers are shown as follows:
* We compare the performance with the Recurrent Entity Networks model (EntNet) BIBREF17
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.