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 |
What are different tools and channels used in organizations for communication and training?
|
There are many different tools and channels used by organizations for communication and training. These range from short and informal, such as a direct message or text, to long and formal, such as a blog article, webinar, or a training class.
Widely use short and informal tools include Slack, Microsoft Team, Skype, WhatsApp, and many more. These tools can incorporate text, audio file, video files, and links to files. They're usually used for one to one, one to many (small group), or one to many (large group).
Email can be used both for short and informal and longer, more formal communications. In fact, emails have fully supplanted the "interoffice memo" used for decades in the corporate world.
Presentations are a widely used tool in enterprises today. These typically include both a live presenter and prepared slides for the audience to see, and usually, take away. Tools for creating these slide show, also referred to as "decks," include Microsoft PowerPoint and Google Slides. Webinars are virtual presentations, nearly always using slides.
An interesting evolution in corporate communication has been the advent of the "slideument." Coined by Nancy Duarte in her book, slide:ology, a slideument is "the worst of both worlds." Each slides features a great deal of text, which may or may not be the presenter's script. However, being slides, there is often little heed given to the principles of good writing. Directness, transitions, conciseness, clarity.
A workshop or training event is another method of corporate communication. The key difference between this and other tools is that a workshop is usually for skill building of some type and incorporates (hopefully) a great deal of audience interaction (with the presenter and with other participants. Slides are often used, but other tools can be brought into play, such as flip charts, small group discussions, and assignments for individuals and groups.
|
|
null | false
| 274
|
Internet “trolls” are users of an online community who quarrel and upset people, seeking to sow discord by posting inflammatory content. More recently, organized “troll farms” of political opinion manipulation trolls have also emerged.
Such farms usually consist of state-sponsored agents who control a set of pseudonymous user accounts and personas, the so-called “sockpuppets”, which disseminate misinformation and propaganda in order to sway opinions, destabilize the society, and even influence elections BIBREF0.
The behavior of political trolls has been analyzed in different recent circumstances, such as the 2016 US Presidential Elections and the Brexit referendum in UK BIBREF0, BIBREF1. However, this kind of analysis requires painstaking and time-consuming manual labor to sift through the data and to categorize the trolls according to their actions. Our goal in the current paper is to automate this process with the help of machine learning (ML). In particular, we focus on the case of the 2016 US Presidential Elections, for which a public dataset from Twitter is available. For this case, we consider only accounts that post content in English, and we wish to divide the trolls into some of the functional categories identified by BIBREF0: left troll, right troll, and news feed.
We consider two possible scenarios. The first, prototypical ML scenario is supervised learning, where we want to learn a function from users to categories {left, right, news feed}, and the ground truth labels for the troll users are available. This scenario has been considered previously in the literature by BIBREF2. Unfortunately, a solution for such a scenario is not directly applicable to a real-world use case. Suppose a new troll farm trying to sway the upcoming European or US elections has just been discovered. While the identities of the accounts might be available, the labels to learn from would not be present. Thus, any supervised machine learning approach would fall short of being a fully automated solution to our initial problem.
A more realistic scenario assumes that labels for troll accounts are not available. In this case, we need to use some external information in order to learn a labeling function. Indeed, we leverage more persistent entities and their labels: news media. We assume a learning scenario with distant supervision where labels for news media are available. By combining these labels with a citation graph from the troll accounts to news media, we can infer the final labeling on the accounts themselves without any need for manual labeling.
One advantage of using distant supervision is that we can get insights about the behavior of a newly-discovered troll farm quickly and effortlessly. Differently from troll accounts in social media, which usually have a high churn rate, news media accounts in social media are quite stable. Therefore, the latter can be used as an anchor point to understand the behavior of trolls, for which data may not be available.
We rely on embeddings extracted from social media. In particular, we use a combination of embeddings built on the user-to-user mention graph, the user-to-hashtag mention graph, and the text of the tweets of the troll accounts. We further explore several possible approaches using label propagation for the distant supervision scenario.
As a result of our approach, we improve the classification accuracy by more than 5 percentage points for the supervised learning scenario. The distant supervision scenario has not previously been considered in the literature, and is one of the main contributions of the paper. We show that even by hiding the labels from the ML algorithm, we can recover 78.5% of the correct labels.
The contributions of this paper can be summarized as follows:
We predict the political role of Internet trolls (left, news feed, right) in a realistic, unsupervised scenario, where labels for the trolls are not available, and which has not been explored in the literature before.
We propose a novel distant supervision approach for this scenario, based on graph embeddings, BERT, and label propagation, which projects the more-commonly-available labels for news media onto the trolls who cited these media.
We improve over the state of the art in the traditional, fully supervised setting, where training labels are available.
We show that even by hiding the labels from the ML algorithm, we can recover 78.5% of the correct labels
|
How much of the correct labels can they recover even by hiding the labels from the ML algorithm?
|
78.5%.
|
null | false
| null |
Name some famous actors in the 1940's
|
Jennifer Jones, Barbara Stanwyck, Greer Garson, Humphrey Bogart, Bette Davis, Bob Hope, Gregory Peck, David Niven, Victor Mature, James Cagney, Betty Grable, Judy Garland, Janet Leigh, Spencer Tracy.
|
|
null | false
| null |
Are all types of salt the same?
|
No, definitely not. There are varying sized grains or crystals of salt, various production methods, different colors, and potentially different additives. For example, a fine-grained iodized salt is highly processed with additives, while a finishing salt may be very unprocessed natural crystals harvested directly from sea water.
|
|
1909.11297
| true
| null |
Secondly, we compare the performance of three BERT-related methods. The performance of BERT-Original and BERT-Soft are similar by comparing their average scores. The reason may be that the original BERT has already modeled the deep relationships between the sentence and the aspect. BERT-Original can be thought of as a kind of soft-selection approach as BERT-Soft. We also observe that the snippet selection by reinforcement learning improves the performance over soft-selection approaches in almost all settings. However, the improvement of BERT-Hard over BERT-Soft is marginal. The average score of BERT-Hard is better than BERT-Soft by 0.68%. The improvement percentages are between 0.36% and 1.49%, while on the Laptop dataset, the performance of BERT-Hard is slightly weaker than BERT-Soft. The main reason is that the datasets only contain a small portion of multi-aspect sentences with different polarities. The distraction of attention will not impact the sentiment prediction much in single-aspect sentences or multi-aspect sentences with the same polarities.
The main reason is that the datasets only contain a small portion of multi-aspect sentences with different polarities. The distraction of attention will not impact the sentiment prediction much in single-aspect sentences or multi-aspect sentences with the same polarities.
|
Is the model evaluated against the baseline also on single-aspect sentences?
|
No.
|
null | false
| null |
What does it mean if a dinner "smacks"?
|
Food "smacks" when it is really good and tasty. It's possibly a reference to "lip-smacking" when food is especially good. When you love what you're eating, you can say "this food smacks".
|
|
null | false
| null |
What kind of species is the Hackelia bella
|
Hackelia bella is a species of flowering plant in the borage family known by the common name greater showy stickseed.
It is native to the northern California Coast Ranges and the Klamath Mountains in northeastern California and southwestern Oregon in the United States. It is found in yellow pine forest, red fir forest habitats.
|
|
null | false
| null |
Write a creative ending scene for a thriller movie
|
After 18 months of finding the treasure, the people of Corinth gathered to praise Big B and his friends for their achievement. Amidst the celebrations, the friends saw a shady person new to the town. The person approached Big B and his friends, asking for a moment to speak. He presented them with a captain's log from 1867 for the friends to look into. The shady man was old and hence asked the friends for their assistance in locating what the captain's log pointed to. And when asked who the captain was, he said the diary belonged to Marshall Teach alias BLACKBEARD - one of the greatest and most vicious pirates to live.
|
|
null | false
| null |
What are the main ingredients for bread?
|
Flour, water, yeast and salt.
|
|
null | false
| null |
What is locus (in genomics)?
|
In genetics and genomics, a locus is a specific, fixed position on a chromosome where a particular gene or genetic marker is located. Each chromosome carries many genes, with each gene occupying a different position or locus; in humans, the total number of protein-coding genes in a complete haploid set of 23 chromosomes is estimated at 19,000–20,000.
Genes may possess multiple variants known as alleles, and an allele may also be said to reside at a particular locus. Diploid and polyploid cells whose chromosomes have the same allele at a given locus are called homozygous with respect to that locus, while those that have different alleles at a given locus are called heterozygous. The ordered list of loci known for a particular genome is called a gene map. Gene mapping is the process of determining the specific locus or loci responsible for producing a particular phenotype or biological trait. Association mapping, also known as "linkage disequilibrium mapping", is a method of mapping quantitative trait loci (QTLs) that takes advantage of historic linkage disequilibrium to link phenotypes (observable characteristics) to genotypes (the genetic constitution of organisms), uncovering genetic associations.
|
|
null | false
| null |
In what country have many of the most important hominid fossils been found?
|
Many important hominid fossils have been found in Tanzania, such as 6-million-year-old Pliocene hominid fossils. The genus Australopithecus ranged across Africa between 4 and 2 million years ago, and the oldest remains of the genus Homo are found near Lake Olduvai. Following the rise of Homo erectus 1.8 million years ago, humanity spread all over the Old World, and later in the New World and Australia under the species Homo sapiens. H. sapiens also overtook Africa and absorbed the older species of humanity. Later in the Stone and Bronze Age, prehistoric migrations into Tanzania included Southern Cushitic speakers who moved south from present-day Ethiopia; Eastern Cushitic people who moved into Tanzania from north of Lake Turkana about 2,000 and 4,000 years ago; and the Southern Nilotes, including the Datoog, who originated from the present-day South Sudan–Ethiopia border region between 2,900 and 2,400 years ago.: page 18 These movements took place at about the same time as the settlement of the Mashariki Bantu from West Africa in the Lake Victoria and Lake Tanganyika areas. They subsequently migrated across the rest of Tanzania between 2,300 and 1,700 years ago.
|
Many hominid fossils have been found inTanzania.
|
null | false
| null |
Classify these countries as being in the northern hemisphere or the southern hemisphere: Brazil, India, Mexico, Australia, New Zealand, South Africa, Canada, United Kingdom
|
The countries in the northern hemisphere are: India, Mexico, Canada, United Kingdom
The countries in the southern hemisphere are: Brazil, Australia, New Zealand, South Africa
|
|
1704.02385
| false
| null |
We collected all available comments in the stories from Reddit from August 2015. Reddit is popular website that allows registered users (without identity verification) to participate in forums specific a post or topic. These forums are of they hierarchical type, those that allow nested conversation, where the children of a comment are its direct response. To increase recall and make the annotation process feasible we created an inverted index with Lucene and queried for comments containing the word troll with an edit distance of 1, to include close variations of this word. We do so inspired by the method by BIBREF2 to created a bullying dataset, and because we hypothesize that such comments will be related or involved in a trolling event. As we observed in the dataset, people use the word troll in many different ways, sometimes it is to point out that some used is indeed trolling him or her or is accusing someone else of being a troll. Other times, people use the term, to express their frustration or dislike about a particular user, but there is no trolling event. Other times, people simple discuss about trolling and trolls, without actually participating or observing one directly. Nonetheless, we found that this search produced a dataset in which 44.3 % of the comments directly involved a trolling event. Moreover, as we exposed our trolling definition, it is possible for commentators in a conversation to believe that they are witnessing a trolling event and respond accordingly even where there is none. Therefore, even in the comments that do not involve trolling, we are interested in learning what triggers users interpretation of trolling where it is not present and what kind of response strategies are used. We define as a suspected trolling event in our dataset a comment in which at least one of its children contains the word troll.
We collected all available comments in the stories from Reddit from August 2015.
|
what is the source of the new dataset?
|
The answers are shown as follows:
* Reddit
|
1906.03338
| false
| null |
The results in Table TABREF38 confirm the results of BIBREF13 and suggest that we successfully replicated a large proportion of their features.
The results for all three prediction settings (one outgoing edge: INLINEFORM0 , support/attack: INLINEFORM1 and support/attack/neither: INLINEFORM2 ) across all type variables ( INLINEFORM3 , INLINEFORM4 and INLINEFORM5 ) are displayed in Table TABREF39 . All models significantly outperform the majority baseline with respect to macro F1. Intriguingly, the content-ignorant models ( INLINEFORM6 ) always perform significantly better than the models which only have access to the EAUs' content ( INLINEFORM7 , INLINEFORM8 ). In the most general task formulation ( INLINEFORM9 ), we observe that INLINEFORM10 even significantly outperforms the model which has maximum access (seeing both EAU spans and surrounding contexts: INLINEFORM11 ).
The results in Table TABREF38 confirm the results of BIBREF13 and suggest that we successfully replicated a large proportion of their features.
The results for all three prediction settings (one outgoing edge: INLINEFORM0 , support/attack: INLINEFORM1 and support/attack/neither: INLINEFORM2 ) across all type variables ( INLINEFORM3 , INLINEFORM4 and INLINEFORM5 ) are displayed in Table TABREF39 . All models significantly outperform the majority baseline with respect to macro F1.
|
What baseline and classification systems are used in experiments?
|
The answers are shown as follows:
* BIBREF13
* majority baseline
|
null | false
| null |
The black swan theory or theory of black swan events is a metaphor that describes an event that comes as a surprise, has a major effect, and is often inappropriately rationalized after the fact with the benefit of hindsight. The term is based on an ancient saying that presumed black swans did not exist – a saying that became reinterpreted to teach a different lesson after they were discovered in Australia.
The theory was developed by Nassim Nicholas Taleb, starting in 2001, to explain:
1. The disproportionate role of high-profile, hard-to-predict, and rare events that are beyond the realm of normal expectations in history, science, finance, and technology.
2. The non-computability of the probability of consequential rare events using scientific methods (owing to the very nature of small probabilities).
3. The psychological biases that blind people, both individually and collectively, to uncertainty and the substantial role of rare events in historical affairs.
Taleb's "black swan theory" refers only to unexpected events of large magnitude and consequence and their dominant role in history. Such events, considered extreme outliers, collectively play vastly larger roles than regular occurrences. More technically, in the scientific monograph "Silent Risk", Taleb mathematically defines the black swan problem as "stemming from the use of degenerate metaprobability".
|
Who developed the Black Swan theory?
|
The black Swan Theory was developed by Nassim Nicholas Taleb in 2001.
|
null | false
| null |
What are hiccups and why do we get them?
|
"Hiccup comes from the Latin phrase "to catch one's breath while sobbing". Hiccups are involuntary contractions of the diaphragm that produces a "hic" sound. The time between these "hics" can be erratic but are usually pretty consistant. Pregnant women have been known to say they can feel their unborn babies hiccuping in the womb.
If you wait long enough hiccups usually just go away on their own.
There are many "Folk cures" for hiccups. These are very varied. There are no real medical remedies. The more people you talk to the more varied these remedies are. Everyone seems to have their own unique remedy.
Some of these include: head standing, drinking a glass of water upside-down, being frightened by someone, breathing into a bag, eating a large spoonful of peanut butter and placing sugar on or under the tongue and my favorite -Drinking through a straw with the ears plugged!
|
|
null | false
| 462
|
Our experiment in Section 4.1 showed that linear models exhibit a very clean trade-off between complexity and generalization. However there are many subtle factors that influence effective model complexity in neural networks besides architecture: learning rate, learning rate schedule, weight decay, early stopping, etc; and the repeatably of learning is hampered by non-convexity, stochastic initialization, and stochastic mini-batches, etc. The previous evaluation study of attempted some degree of control over these factors for state of the art neural DG methods by applying a common hyperparameter search strategy that aimed to find the best tuning (cf: Fig) for each competitor. They nevertheless found existing methods to perform unreliably. Our conjecture is that the number of seed samples (3) and hyper-parameter search trials (20) used was far too few to reliably tune the deep networks evaluated, and therefore the tuning of each model was inaccurate and performance estimates unreliable. Since dense and accurate hyper-parameter tuning for neural networks is more challenging than our linear models in Section 4.1, we take a different approach. Rather than trying to control complexity directly, we take trained DomainBed methods and measure their complexity retrospectively.
Setup To ensure that complexity can be measured accurately, we work with 2-layer MLPs. Specifically, we take fixed ImageNet pre-trained ResNet-18 features, and feed them to MLPs, which are then trained using the DomainBed framework. 2-layer MLPs are sufficient to instantiate many of the state of the art DG algorithms in DomainBed. We train and compare ERM, CORAL, Mixup , MMD, RSC, SD, VRex, and IRM with two hyperparameter choices. We also report the results of an ERM model, checkpointed at a range of training iterations.
Measuring Model Complexity Because we have limited our attention to 2-layer MLPs, we can take advantage of a model capacity metric that is specialised for this class of models. We retrospectively determine the complexity of a trained network using the measure proposed by Neyshabur et al. (), who also use this measure to bound the Rademacher complexity of a 2-layer MLP hypothesis classes. More concretely, the expression used for computing complexity is
where U is the weight matrix of the first layer, U 0 is its random initialisation, and V is the weight matrix of the second layer. We use • F to denote Frobenius norm, and • 2 to indicate the spectral norm. Note that for simplicity we have omitted constant factors that depend only on the architecture and problem setting, and not the learned weights, as we use the same architecture for all methods we investigate.
The results in Figure summarise the trade-off between measured model complexity (xaxis), and held-out domain test accuracy (y-axis), averaged over all choices of held out domain for each dataset (See Supplementary for full breakdown across all held-out domains). The top plot compares several of the published neural DG methods implemented in DomainBed. The main message of the top plot is that results are consistent with the hypothesis that the resulting complexity of the models trained by various DG methods is a key factor in determining domain generalisation accuracy (compare Fig. with the clean result for linear where models we are able to intervene and control complexity directly in Fig.)-thus explaining the previously erratic behaviour observed in domain-bed.
To provide a different view of the same issue, the bottom plot reports a vanilla untuned ERM model check-pointed every 300 training iterations between (up to a total of 15,000) but expressed in the same complexity units as above. Because neural models can gain complexity with iterations (see, e.g.,;) we also see typical overfitting-underfitting trade-off curves. This shows that, as expected, proper choice of early stopping criterion is important. It also shows that over a similar dynamic range of complexity (0.5-3), ERM (below) and alternative models (above) span a similar dynamic range of accuracy (e.g., 6% for PACS). This suggests that complexity, as controlled by whatever mechanism, is a key determinant of DG performance.
How Do DG Models Control Complexity? By introducing different modifications to standard ERM, DG models explicitly or implicitly modify the bias and variance of the function class to be learned. highlight neural models as being dependent on learning rate, batch size, dropout rate, and weight decay; with other factors being choice of optimiser and weight initialiser, etc. For example, higher dropout rate and weight decay tend to reduce complexity, while some other factors influence complexity in a less transparent way. RSC introduce more sophisticated dropout-like mechanisms, which would be expected to reduce complexity. Meanwhile alignment-based methods like CORAL and MMD effectively add auxiliary losses, which will implicitly affect complexity, but are hard to explicitly link to it. Consistency based methods like IRM and VRex penalise loss variance, which also tends to reduce the generalisation gap in the single task case. The specific setting of all the corresponding hyper-parameters (e.g., regulariser strength, dropout rates) influence final model complexity, which we argue is the key determinant of performance.
Figure 2: Performance of neural networks as a function of their measured model complexity after training using DomainBed (DB). The overall results are consistent with a standard bias-variance trade-off (cf. Fig 1 for linear models): performance depends on how well each model was tuned by DB’s hyper-parameter search procedure. Top: Leave-one-domain-out cross-validation performance of various neural DG algorithms evaluated by DomainBed. Horizontal error bars correspond to the standard deviation of the model complexity measured for each cross-validation iteration. Bottom: Performance of ERM model checkpointed at different training iterations. The central tendency is obtained via fitting a support vector regression model (Shevade et al., 2000) with a 6th order polynomial kernel.
|
Is there any clear trait of the trade-off in Fig (2)?How about the IRM Results in Fig 2?
|
We now illustrate the tradeoff between overfitting and underfitting regimes using polynomial regression curves. In the underfitting regime (left) performance is sharply lower, in the overfitting regime (right) performance decays slowly, with best performance somewhere in the middle. With regards to IRM: Short answer: The apparent discrepancy is not due to MLP vs ResNet and we do not contradict the prior work. We now include a version of IRM that is tuned to perform closer to ERM by reducing its regularisation strength. Long answer:It’s important to note that these models are run without careful hyperparameter tuning in Fig 2. Depending on hyperparameters, models will achieve different complexities (e.g., as we showed in Fig 1). So the specific placement of any given model (such as IRM, ERM, etc) on the graph will depend on hyperparameter choice. Our point is not about the absolute position of any given model, but the fact that when complexity is allowed to vary by any means (due to regulariser strength, training iterations, choice of ERM vs IRM vs REX etc), the resulting complexity (x-axis) determines the performance (y-axis).
|
null | false
| 306
|
All experiments are implemented on the hardware with Intel(R) Xeon(R) CPU E5-2682 v4 @ 2.50GHz and NVIDIA Tesla P100.
We evaluate our method on the SKE dataset used in this competition, which is the largest schema-based Chinese information extraction dataset in the industry, containing more than 430,000 SPO triples in over 210,000 real-world Chinese sentences, bounded by a pre-specified schema with 50 types of predicates. All sentences in SKE Dataset are extracted from Baidu Baike and Baidu News Feeds. The dataset is divided into a training set (170k sentences), a development set (20k sentences) and a testing set (20k sentences). The training set and the development set are to be used for training and are available for free download. The test set is divided into two parts, the test set 1 is available for self-verification, the test set 2 is released one week before the end of the competition and used for the final evaluation.
The max sequence length is set to 128, the number of fully connected layer of relation classification branch is set to 2, and that of global relation branch is set to 1. During training, we use Adam with the learning rate of 2e-5, dropout probability of 0.1. This model converges in 3 epoch.
All uppercase letters are converted to lowercase letters. We use max sequence length 128 so sentences longer than 128 are split by punctuation. According to FAQ, entities in book title mark should be completely extracted. Because the annotation criteria in trainset are diverse, we revise the incomplete entities. To keep consistence, book title marks around the entities are removed.
Our postprocessing mechanism is mainly based on the FAQ evaluation rules. After model prediction, we remove triplets whose entity-relation types are against the given schemas. For entities contained in book title mark, we complement them if they are incomplete. Date type entities are also complemented to the finest grain. These are implemented by regular expression matching.
Note that entity related preprocessing and postprocessing are also performed on the development set to keep consistency with the test set, thus the change of development metric is reliable.
We evaluate our method on the SKE dataset used in this competition, which is the largest schema-based Chinese information extraction dataset in the industry, containing more than 430,000 SPO triples in over 210,000 real-world Chinese sentences, bounded by a pre-specified schema with 50 types of predicates.
|
What is the SKE dataset?
|
It is the largest schema-based Chinese information extraction dataset in the industry, containing more than 430,000 SPO triples in over 210,000 real-world Chinese sentences, bounded by a pre-specified schema with 50 types of predicates.
|
null | false
| 19
|
In this paper we described our system for the CoNLL–SIGMORPHON 2018 shared task on Universal Morphological Reinflection, Task 2, which achieved the best performance out of all systems submitted, an overall accuracy of 49.87. We showed in an ablation study that this is due to three core innovations, which extend a character-based encoder-decoder model: (1) a wide context window, encoding the entire available context; (2) multi-task learning with the auxiliary task of MSD prediction, which acts as a regulariser; (3) a multilingual approach, exploiting information across languages. In future work we aim to gain better understanding of the increase in variance of the results introduced by each of our modifications and the reasons for the varying effect of multi-task learning for different languages.
In future work we aim to gain better understanding of the increase in variance of the results introduced by each of our modifications and the reasons for the varying effect of multi-task learning for different languages.
|
Will the team focus on analyzing the reasons for the varying effect of multi-task learning for different languages in their future work?
|
Yes.
|
null | false
| null |
Which is considered the first human technology?
|
Fire
|
|
null | false
| null |
Identify which car manufacturer is British or American: Jaguar, GMC
|
GMC is American, Jaguar is British
|
|
null | false
| null |
Will, a 45-yr-old former collegiate swimmer turned lifelong triathlete, requests assistance with run training. His only significant medical history is a series of overuse injuries to his shoulders and Achilles tendon. In recent weeks, he notes his workouts are unusually difficult and reports feeling constriction in his chest with exertion – something he attributes to deficiencies in core strength. Upon further questioning, he explains that the chest constriction is improved with rest and that he often feels dizzy during recovery. What do you think is the underlying problem that could be causing Will’s symptoms during/after exercise?
|
Will probably has some form of cardiovascular disease (CVD). As he was a collegiate athlete and now triathlete, he has undergone many years of intense exercise training, which could induce pathological cardiac remodeling, ultimately leading to myocardial fibrosis. This form of CVD can be common among trained athletes, particularly endurance athletes such as Will (swimming, triathlon).
|
|
null | false
| 328
|
Neural network based models have been widely exploited with the prosperities of Deep Learning BIBREF0 and achieved inspiring performances on many NLP tasks, such as text classification BIBREF1 , BIBREF2 , semantic matching BIBREF3 , BIBREF4 and machine translation BIBREF5 . These models are robust at feature engineering and can represent words, sentences and documents as fix-length vectors, which contain rich semantic information and are ideal for subsequent NLP tasks.
One formidable constraint of deep neural networks (DNN) is their strong reliance on large amounts of annotated corpus due to substantial parameters to train. A DNN trained on limited data is prone to overfitting and incapable to generalize well. However, constructions of large-scale high-quality labeled datasets are extremely labor-intensive. To solve the problem, these models usually employ a pre-trained lookup table, also known as Word Embedding BIBREF6 , to map words into vectors with semantic implications. However, this method just introduces extra knowledge and does not directly optimize the targeted task. The problem of insufficient annotated resources is not solved either.
Multi-task learning leverages potential correlations among related tasks to extract common features, increase corpus size implicitly and yield classification improvements. Inspired by BIBREF7 , there are a large literature dedicated for multi-task learning with neural network based models BIBREF8 , BIBREF9 , BIBREF10 , BIBREF11 . These models basically share some lower layers to capture common features and further feed them to subsequent task-specific layers, which can be classified into three types:
In this paper, we propose a generalized multi-task learning architecture with four types of recurrent neural layers for text classification. The architecture focuses on Type-III, which involves more complicated interactions but has not been researched yet. All the related tasks are jointly integrated into a single system and samples from different tasks are trained in parallel. In our model, every two tasks can directly interact with each other and selectively absorb useful information, or communicate indirectly via a shared intermediate layer. We also design a global memory storage to share common features and collect interactions among all tasks.
We conduct extensive experiments on five benchmark datasets for text classification. Compared to learning separately, jointly learning multiple relative tasks in our model demonstrate significant performance gains for each task.
Our contributions are three-folds:
We also design a global memory storage to share common features and collect interactions among all tasks.
|
What is the function of the global memory storage designed by authors?
|
It is used to share common features and collect interactions among all tasks.
|
null | false
| 328
|
Neural network based models have been widely exploited with the prosperities of Deep Learning BIBREF0 and achieved inspiring performances on many NLP tasks, such as text classification BIBREF1 , BIBREF2 , semantic matching BIBREF3 , BIBREF4 and machine translation BIBREF5 . These models are robust at feature engineering and can represent words, sentences and documents as fix-length vectors, which contain rich semantic information and are ideal for subsequent NLP tasks.
One formidable constraint of deep neural networks (DNN) is their strong reliance on large amounts of annotated corpus due to substantial parameters to train. A DNN trained on limited data is prone to overfitting and incapable to generalize well. However, constructions of large-scale high-quality labeled datasets are extremely labor-intensive. To solve the problem, these models usually employ a pre-trained lookup table, also known as Word Embedding BIBREF6 , to map words into vectors with semantic implications. However, this method just introduces extra knowledge and does not directly optimize the targeted task. The problem of insufficient annotated resources is not solved either.
Multi-task learning leverages potential correlations among related tasks to extract common features, increase corpus size implicitly and yield classification improvements. Inspired by BIBREF7 , there are a large literature dedicated for multi-task learning with neural network based models BIBREF8 , BIBREF9 , BIBREF10 , BIBREF11 . These models basically share some lower layers to capture common features and further feed them to subsequent task-specific layers, which can be classified into three types:
In this paper, we propose a generalized multi-task learning architecture with four types of recurrent neural layers for text classification. The architecture focuses on Type-III, which involves more complicated interactions but has not been researched yet. All the related tasks are jointly integrated into a single system and samples from different tasks are trained in parallel. In our model, every two tasks can directly interact with each other and selectively absorb useful information, or communicate indirectly via a shared intermediate layer. We also design a global memory storage to share common features and collect interactions among all tasks.
We conduct extensive experiments on five benchmark datasets for text classification. Compared to learning separately, jointly learning multiple relative tasks in our model demonstrate significant performance gains for each task.
Our contributions are three-folds:
All the related tasks are jointly integrated into a single system and samples from different tasks are trained in parallel.
|
Are the samples from different tasks trained in parallel in the architecture?
|
Yes, they are.
|
null | false
| null |
What is the future trend of job industry
|
More of cloud Machine learning AI security dive into these industries to be on top notch
|
|
null | false
| null |
Should I visit Chicago?
|
Chicago is one of the largest cities in the United States situated on Lake Michigan in the Midwest region of the country. It is a popular travel destination known for its architecture including the world famous Willis Tower (formerly the Sears Tower), the iconic John Hancock Center and the neo-Gothic Tribune Tower. It is also home to some world famous museums including the Art Institute of Chicago and several sports teams such as the Chicago Bulls, Chicago Bears and Chicago Cubs.
|
|
null | false
| null |
What are all the different evolutions for Eevee in Pokemon GO?
|
Eevee has eight different evolutions including Espeon, Vaporeon, Flareon, Jolteon, Leafeon, Glaceon, Umbreon and Sylveon.
|
|
null | false
| null |
Give me the answer to this riddle, explain the answer, and explain why the answer may be challenging to identify: what gets wetter as it dries?
|
The answer to this riddle is a towel. A towel gets wetter when it is used to dry something off. This answer may be challenging to determine because of the contradictory nature of the qualities the riddle is asking for - drying and getting wetter.
|
|
null | false
| null |
Identify which instrument is string or percussion: Talempong, Guitar
|
Guitar is string, Talempong is percussion.
|
|
null | false
| null |
James Leslie Mitchell gained attention from his earliest attempts at fiction, notably from H. G. Wells, but it was his trilogy entitled A Scots Quair, and in particular its first book Sunset Song, with which he made his mark. A Scots Quair, with its combination of stream-of-consciousness, lyrical use of dialect, and social realism, is considered to be among the defining works of the 20th century Scottish Renaissance. It tells the story of Chris Guthrie, a young woman growing up in the north-east of Scotland in the early 20th century. All three parts of the trilogy have been turned into serials by BBC Scotland, written by Bill Craig, with Vivien Heilbron as Chris. Additionally, Sunset Song has been adapted into a film, released in 2015. Spartacus, a novel set in the famous slave revolt, is his best-known full-length work outside this trilogy.
|
Extract all of the names of people mentioned in this paragraph and list them using bullets in the format {Name}
|
• James Leslie Mitchell
• H. G. Wells
• Chris Guthrie
• Bill Craig
• Vivien Heilbron
|
null | false
| null |
Who created pilates?
|
Pilates was developed by Joseph Pilates.
|
|
null | false
| 334
|
Automated deriving discourse relation from free text is a challenging but im-portant problem. The shallow discourse parsing is very useful in the text summariza-tion BIBREF0 , opinion analysis BIBREF1 and natural language generation. Shallow discourse parser is the system of parsing raw text into a set of discourse relations between two adjacent or non-adjacent text spans. Discourse relation is composed of a discourse connective, two arguments of the discourse connective and the sense of the discourse connective. Discourse connective signals the explicit dis-course relation, but in non-explicit discourse relation, a discourse connective is omit-ted. Two arguments of the discourse connective, Arg1 and Arg2, which are the two adjacent or non-adjacent text spans connecting in the discourse relation. The sense of the discourse connective characterizes the nature of the discourse relations. The following discourse relation annotation is taken from the document in the PDTB. Arg1 is shown in italicized, and Arg2 is shown in bold. The discourse connective is underlined.
The connective identifier finds the connective word, “unless”. The arguments identifier locates the two arguments of “unless”. The sense classifier labels the dis-course relation. The non-explicit identifier checks all the pair of adjacent sentences. If the non-explicit identifier indentifies the pair of sentences as non-explicit relation, it will label it the relation sense. Though many research work BIBREF2 , BIBREF3 , BIBREF4 are committed to the shallow discourse parsing field, all of them are focus on the subtask of parsing only rather than the whole parsing process. Given all that, a full shallow discourse parser framework is proposed in our paper to turn the free text into discourse relations set. The parser includes connective identifier, arguments identifier, sense classifier and non-explicit identifier, which connects with each other in pipeline. In order to enhance the performance of the parser, the feature-based maximum entropy model approach is adopted in the experiment. Maximum entropy model offers a clean way to combine diverse pieces of contextual evidence in order to estimate the probability of a certain linguistic class occurring with a certain linguistic context in a simple and accessible manner. The three main contributions of the paper are:
The rest of this paper is organized as follows. Section 2 reviews related work in discourse parsing. Section 3 describes the experimental corpus–PDTB. Section 4 de-scribes the framework and the components of the parser. Section 5 presents experi-ments and evaluations. Conclusions are presented in the Section 6.
Maximum entropy model offers a clean way to combine diverse pieces of contextual evidence in order to estimate the probability of a certain linguistic class occurring with a certain linguistic context in a simple and accessible manner
|
What can be done by the maximum entropy model?
|
It offers a clean way to combine diverse pieces of contextual evidence in order to estimate the probability of a certain linguistic class occurring with a certain linguistic context in a simple and accessible manner.
|
null | false
| null |
The Panama Canal is an artificial 82 km (51 mi) waterway in Panama that connects the Atlantic Ocean with the Pacific Ocean and divides North and South America. The canal cuts across the Isthmus of Panama and is a conduit for maritime trade. One of the largest and most difficult engineering projects ever undertaken, the Panama Canal shortcut greatly reduces the time for ships to travel between the Atlantic and Pacific oceans, enabling them to avoid the lengthy, hazardous Cape Horn route around the southernmost tip of South America via the Drake Passage or Strait of Magellan.
Colombia, France, and later the United States controlled the territory surrounding the canal during construction. France began work on the canal in 1881, but stopped because of lack of investors' confidence due to engineering problems and a high worker mortality rate. The United States took over the project on May 4, 1904, and opened the canal on August 15, 1914. The US continued to control the canal and surrounding Panama Canal Zone until the 1977 Torrijos–Carter Treaties provided for its handover to Panama. After a period of joint American–Panamanian control, the canal was taken over by the Panamanian government in 1999. It is now managed and operated by the government-owned Panama Canal Authority.
Canal locks at each end lift ships up to Gatun Lake, an artificial lake 26 meters (85 ft) above sea level, created to reduce the amount of excavation work required for the canal, and then lower the ships at the other end. The original locks are 33.5 meters (110 ft) wide. A third, wider lane of locks was constructed between September 2007 and May 2016. The expanded waterway began commercial operation on June 26, 2016. The new locks allow transit of larger, New Panamax ships.
Annual traffic has risen from about 1,000 ships in 1914, when the canal opened, to 14,702 vessels in 2008, for a total of 333.7 million Panama Canal/Universal Measurement System (PC/UMS) tons. By 2012, more than 815,000 vessels had passed through the canal. In 2017 it took ships an average of 11.38 hours to pass between the canal's two locks. The American Society of Civil Engineers has ranked the Panama Canal one of the seven wonders of the modern world.
|
From this paragraph, find out who built the Panama Canal
|
France began work on the Panama canal in 1881, but stopped because of lack of investors' confidence due to engineering problems and a high worker mortality rate. The United States took over the project on May 4, 1904, and opened the canal on August 15, 1914.
|
null | false
| null |
Who is the 6th player in Women's NCAA D-I Basketball to score 1,000 points in a single season?
|
If you watched any of the Women's March Madness 2023 games, especially Iowa, then you witnessed greatness in the making by Caitlin Clark. This 1,000 point feat in a single season, happened midway through the game against the powerhouse (undefeated at the time) South Carolina Gamecocks on March 31, 2023. It's also worth noting that in the same game, Caitlin dropped 41 points and her three-point range is Steph Curry like.
|
|
null | false
| null |
Identify which animal species is alive or extinct: Quagga, Beluga
|
Beluga is alive, Quagga is extinct.
|
|
null | false
| null |
Why has pickleball become so popular?
|
Pickleball has indeed become very popular. Promoters of the game have been calling it the "fastest growing sport" for some time now.
There are a few reasons many people cite for why the sport is so popular:
- It is a quick game to learn. The rules are pretty simple, and more importantly, the skill needed to hit functional shots in the game is not very difficult to learn. Compare it to tennis, where it can take multiple lessons from a professional teacher just to get to a point where you can hit a few good shots. With pickleball, almost anyone of any age or fitness level can pick up a paddle and be playing a game in just minutes.
- Pickleball culture has tended to be very inclusive. Many pickleball groups make it a point to be very inclusive of all ages, races and backgrounds. This often creates a sense of community for players, and makes them more likely to keep playing and invite friends along, thus growing the sport.
- It is a fun way to get in a decent workout. While not as taxing as, say, tennis, pickleball can be a good workout. Especially since unlike tennis there is very little downtime. The pace of games is very quick and involves near constant movement. The workout is not cardio-intensive, but still it is an easy way to have fun while staying active.
|
|
null | false
| null |
Which of these cities in the United States of America are the capital of their state: Albany, Albuquerque, Anchorage, Ann Arbor, Annapolis?
|
Albany is the capital of New York and Annapolis is the capital of Maryland. The other cities are not the capitals of their state: Albuquerque is a city in New Mexico but is not the capital, Anchorage is a city in Alaska but is not the capital, and Ann Arbor is a city in Michigan but is not the capital.
|
|
null | false
| null |
The 2023 NCAA Division I men's basketball tournament involved 68 teams playing in a single-elimination tournament that determined the National Collegiate Athletic Association (NCAA) Division I men's basketball national champion for the 2022–23 season. The 84th annual edition of the tournament began on March 14, 2023, and concluded with the UConn Huskies defeating the San Diego State Aztecs, 76–59 in the championship game on April 3 at NRG Stadium in Houston, Texas.
|
Who won March Madness?
|
UConn Huskies won the 2023 NCAA Division 1 men's basketball tournament defeating San Diego State Aztecs, 76-59. This tournament is popularly known as March Madness.
|
null | false
| 114
|
Dialogue evaluation is an open problem and existing works have found that automatic metrics have low correlation to human evaluation BIBREF35 , BIBREF36 , BIBREF37 . Thus, we resorted to manual evaluation to assess the generation quality on WeiboDial. We randomly sampled 200 posts from the test set and collected the generated results from all the models. For each pair of responses (one from ARAML and the other from a baseline, given the same input post), five annotators were hired to label which response is better (i.e. win, lose or tie) in terms of grammaticality (whether a response itself is grammatical and logical) and relevance (whether a response is appropriate and relevant to the post). The two metrics were evaluated independently.
The evaluation results are shown in Table TABREF35 . To measure the inter-annotator agreement, we calculated Fleiss' kappa BIBREF38 for each pair-wise comparison where results show moderate agreement ( INLINEFORM0 ). We also conducted sign test to check the significance of the differences.
As shown in Table TABREF35 , ARAML performs significantly better than other baselines in all the cases. This result indicates that the samples surrounding true responses provide stable rewards for the generator, and stable RAML training paradigm significantly enhances the performance in both metrics.
For each pair of responses (one from ARAML and the other from a baseline, given the same input post), five annotators were hired to label which response is better (i.e. win, lose or tie) in terms of grammaticality (whether a response itself is grammatical and logical) and relevance (whether a response is appropriate and relevant to the post). The two metrics were evaluated independently.
|
What are the two metrics of dialogue evaluation?
|
They are grammaticality and relevance.
|
null | false
| null |
Who wrote the Lord of the Rings?
|
Lord of the Rings was written by J.R.R. Tolkien. Tolkien also wrote The Hobbit.
|
|
null | false
| null |
Zygmuntówek [zɨɡmunˈtuvɛk] is a village in the administrative district of Gmina Malanów, within Turek County, Greater Poland Voivodeship, in west-central Poland. It lies approximately 2 kilometres (1 mi) north of Malanów, 10 km (6 mi) south-west of Turek, and 112 km (70 mi) south-east of the regional capital Poznań.
|
What direction (North-South-East-West) is Malanów from Zygmuntówek?
|
Malanów is south of Zygmuntówek. We know this because Zygmuntówek is north of Malanów.
|
null | false
| 57
|
This work has applied a multiscale graph partitioning algorithm (Markov Stability) to extract content-based clusters of documents from a textual dataset of healthcare safety incident reports in an unsupervised manner at different levels of resolution. The method uses paragraph vectors to represent the records and obtains an ensuing similarity graph of documents constructed from their content. The framework brings the advantage of multi-resolution algorithms capable of capturing clusters without imposing a priori their number or structure. Since different levels of resolution of the clustering can be found to be relevant, the practitioner can choose the level of description and detail to suit the requirements of a specific task.
Our a posteriori analysis evaluating the similarity against the hand-coded categories and the intrinsic topic coherence of the clusters showed that the method performed well in recovering meaningful categories. The clusters of content capture topics of medical practice, thus providing complementary information to the externally imposed classification categories. Our analysis shows that some of the most relevant and persistent communities emerge because of their highly homogeneous medical content, although they are not easily mapped to the standardised external categories. This is apparent in the medically-based content clusters associated with Labour ward, Pressure ulcer, Chemotherapy, Radiotherapy, among others, which exemplify the alternative groupings that emerge from free text content.
The categories in the top level (Level 1) of the pre-defined classification hierarchy are highly diverse in size (as shown by their number of assigned records), with large groups such as `Patient accident', `Medication', `Clinical assessment', `Documentation', `Admissions/Transfer' or `Infrastructure' alongside small, specific groups such as `Aggressive behaviour', `Patient abuse', `Self-harm' or `Infection control'. Our multi-scale partitioning finds corresponding groups in content across different levels of resolution, providing additional subcategories with medical detail within some of the large categories (as shown in Fig. 4 and S1 ). An area of future research will be to confirm if the categories found by our analysis are consistent with a second level in the hierarchy of external categories (Level 2, around 100 categories) that is used less consistently in hospital settings. The use of content-driven classification of reports could also be important within current efforts by the World Health Organisation (WHO) under the framework for the International Classification for Patient Safety (ICPS) BIBREF48 to establish a set of conceptual categories to monitor, analyse and interpret information to improve patient care.
One of the advantages of a free text analytical approach is the provision, in a timely manner, of an intelligible description of incident report categories derived directly from the rich description in the 'words' of the reporter themselves. The insight from analysing the free text entry of the person reporting could play a valuable role and add rich information than would have otherwise been obtained from the existing approach of pre-defined classes. Not only could this improve the current state of play where much of the free text of these reports goes unused, but it avoids the fallacy of assigning incidents to a pre-defined category that, through a lack of granularity, can miss an important opportunity for feedback and learning. The nuanced information and classifications extracted from free text analysis thus suggest a complementary axis to existing approaches to characterise patient safety incident reports.
Currently, local incident reporting system are used by hospitals to submit reports to the NRLS and require risk managers to improve data quality of reports, due to errors or uncertainty in categorisation from reporters, before submission. The application of free text analytical approaches, like the one we have presented here, has the potential to free up risk managers time from labour-intensive tasks of classification and correction by human operators, instead for quality improvement activities derived from the intelligence of the data itself. Additionally, the method allows for the discovery of emerging topics or classes of incidents directly from the data when such events do not fit the pre-assigned categories by using projection techniques alongside methods for anomaly and innovation detection.
In ongoing work, we are currently examining the use of our characterisation of incident reports to enable comparisons across healthcare organisations and also to monitor their change over time. This part of ongoing research requires the quantification of in-class text similarities and to dynamically manage the embedding of the reports through updates and recalculation of the vector embedding. Improvements in the process of robust graph construction are also part of our future work. Detecting anomalies in the data to decide whether newer topic clusters should be created, or providing online classification suggestions to users based on the text they input are some of the improvements we aim to add in the future to aid with decision support and data collection, and to potentially help fine-tune some of the predefined categories of the external classification.
An area of future research will be to confirm if the categories found by our analysis are consistent with a second level in the hierarchy of external categories (Level 2, around 100 categories) that used less consistently in hospital settings.
|
What will be confirmed as an area of future research?
|
An area of future research will be to confirm if the categories found by their analysis are consistent with a second level in the hierarchy of external categories (Level 2, around 100 categories) that used less consistently in hospital settings.
|
null | false
| null |
What is IAAS?
|
IAAS is an acronym for Infrastructure-as-a-service and refers to the networking, storage, compute and virtualisation services the form the foundation of public cloud computing.
|
|
null | false
| null |
Give me a bunch of ways that you can get information from a book.
|
There a number of ways to gather information from a book, you can listen to the audiobook version, there are a number of services that provide abridged versions of books that you can try, alternatively you can simply read the book.
|
|
null | false
| null |
Pompeii (/pɒmˈpeɪ(i)/, Latin: [pɔmˈpei̯.iː]) was an ancient city located in what is now the comune of Pompei near Naples in the Campania region of Italy. Pompeii, along with Herculaneum and many villas in the surrounding area (e.g. at Boscoreale, Stabiae), was buried under 4 to 6 m (13 to 20 ft) of volcanic ash and pumice in the Eruption of Mount Vesuvius in 79 AD.
|
Where is Pompeii located?
|
The ancient city of Pompeii is located near Naples in the Campania region of Italy.
|
null | false
| null |
Who does Machiavelli believe exemplifies the characteristics of a true prince in The Prince, and why does he hold this belief?
|
In Machiavelli’s, The Prince, Machiavelli describes in great detail the extent to which a true prince should rule their state to ensure complete dominance and obedience of its people. He explains the numerous prerequisites and responsibilities of a prince, in which he exemplifies throughout the text through the different figures, in both positive and negative contexts. Throughout The Prince, one figure, who Machiavelli makes clear he has great respect for, highlighting quite frequently, is Cesare Borgia whom he even refers to with the status comparable to the level of God exclaiming that “there was one man who showed glimpses of greatness... that made you think he was sent by God” (26). Similar rhetoric is frequently used to describe Cesare Borgia throughout the text. In search of who truly exemplifies the characterization of “The Prince,” it can generally seem to be Cesare Borgia from the high praise of Machiavelli, but in reality this is not the case. Rather, the title of “The Prince” belongs to a more subtle figure: Ferdinand of Aragon, the King of Spain. This connection is made because of the numerous contradictions Machiavelli makes in praise of Cesare Borgia and other important figures in the text, while many key characteristics of a true prince that Machiavelli speaks of hold without contradiction for Ferdinand.
Machiavelli exemplifies his teachings through Cesare Borgia’s actions seen in chapter 7 when Cesare Borgia conquered the state of Romagna: immediately crushing the opponent using great cruelty, using a scapegoat to thwart the hatred away from him, and finally killing the scapegoat to create a sense of power and fear within the new states inhabitants. These actions, that explain why he is so highly idolized by Machiavelli, do not go without the fact that other
fundamental characteristics of Cesare Borgia directly contradict Machiavelli's notion of the most dominant prince. A prime example is in how “Cesare Borgia... acquired his state during the ascendancy of his father, and on its decline he lost it” (7). Inheritance obviously requires significantly less effort compared to creating a new principality. Although Cesare Borgia later conquers new principalities, it can be inferred that a rise to power through his fathers fortune was a significant catalyst and does not exemplify virtues of the ideal prince.
Other figures in the book receive similar praise in leadership abilities, but still lack what qualities compare to the ideal prince. These figures include Pope Julius II, whose power is highlighted by Machiavelli in chapter 13 as being consistent more with fortune rather than virtù in his campaign for conquer, Pope Alexander VI, who although displays cunning ability to bring the catholic empire eventual dominance and power, did so with the initial help of being “granted a title and income from the Roman Catholic Church by his uncle Alonso”.1 Many figures follow the same pattern in that they follow many principles exemplified in The Prince, but of course with so many requisites to be a true prince, these figures do not follow the explicit definition of a prince. One figure however, follows Machiavelli’s points very closely and exemplifies the fundamentals of a true prince in his beginnings, leaving almost no flaws in his rise to becoming a true prince: Ferdinand of Aragon.
Ferdinand’s characteristics lead him to the title of “The Prince” but a core virtù is what truly sets him apart from the other figures: his rise to power. Almost immediately in The Prince, Machiavelli describes the differences between principalities and emphasizes how maintaining a
1 “Alexander VI.” Encyclopædia Britannica, Encyclopædia Britannica, Inc., www.britannica.com/biography/Alexander-VI.
new principality is far more difficult than inheritance. This is largely evident in Ferdinand’s rise to leadership in what Machiavelli describes him as “, by fame and glory, from being an insignificant king to be the foremost king in Christendom” (21). Ferdinand demonstrates his cunning deception through his plot for power, “always using religion as a plea, so as to undertake greater schemes, himself with pious cruelty to driving out and clearing his kingdom of the Moors” (21). Machiavelli marks this never being “a more admirable example, nor one more rare” in which he uses religion as a cloak of invisibility to deceptively conquer states (21). These natural abilities are what make Ferdinand admirable and what separate him from the rest of the figures in the text. This feeling is shared by Machiavelli exclaiming “that his actions have arisen in such a way, one out of the other, that men have never been given time to work steadily against him” (21).
|
|
null | false
| null |
Caribbean and Central American Action was a 501(c)(3) non-profit organization which has hosted the annual Miami Conference on the Caribbean & Central America for 33 years.
The organization has a multinational Board of Trustees, currently headed by Cornelius B. Prior, Jr., and is based in Washington, DC.
Among the public figures speaking at the 33rd conference, held in 2009 in Miami, were the Prime Minister of Haiti, Jean-Max Bellerive, Canada's Minister of State of Foreign Affairs, Peter Kent and Puerto Rico's Secretary of State, Kenneth McClintock. The organization's policy-oriented programs have focused on tourism, financial services, policy and regulatory environments for business in the Americas, energy security, maritime port security, disaster mitigation, customs, telecommunications, intellectual property rights, information technology, agriculture and telecommunications. Cabinet-level officials have represented the United States at prior conferences
The organization has closed
|
What was the Caribbean and Central America Action
|
Caribbean and Central American Action was a non-profit organization which has hosted the annual Miami Conference on the Caribbean & Central America for 33 years.
The organization has a multinational Board of Trustees, currently headed by Cornelius B. Prior, Jr., and is based in Washington, DC.
Among the public figures speaking at the 33rd conference, held in 2009 in Miami, were the Prime Minister of Haiti, Jean-Max Bellerive, Canada's Minister of State of Foreign Affairs, Peter Kent and Puerto Rico's Secretary of State, Kenneth McClintock. The organization's policy-oriented programs have focused on tourism, financial services, policy and regulatory environments for business in the Americas, energy security, maritime port security, disaster mitigation, customs, telecommunications, intellectual property rights, information technology, agriculture and telecommunications. Cabinet-level officials have represented the United States at prior conferences.
The organization has closed.
|
null | false
| null |
Which is a species of fish? Noodlefish or Noodle
|
Noodlefish
|
|
1708.00214
| false
| null |
We experiment with small feed-forward networks for four diverse NLP tasks: language identification, part-of-speech tagging, word segmentation, and preordering for statistical machine translation.
We experiment with small feed-forward networks for four diverse NLP tasks: language identification, part-of-speech tagging, word segmentation, and preordering for statistical machine translation.
|
What NLP tasks do the authors evaluate feed-forward networks on?
|
The answers are shown as follows:
* language identification, part-of-speech tagging, word segmentation, and preordering for statistical machine translation
|
null | false
| 267
|
Social media is a very popular way for people to express their opinions publicly and to interact with others online. In aggregation, social media can provide a reflection of public sentiment on various events. Unfortunately, many users engaging online, either on social media, forums or blogs, will often have the risk of being targeted or harassed via abusive language, which may severely impact their online experience and the community in general. The existence of social networking services creates the need for detecting user-generated hateful messages prior to publication. All published text that is used to express hatred towards some particular group with the intention to humiliate its members is considered a hateful message.
Although hate speech is protected under the free speech provisions in the United States, there are other countries, such as Canada, France, United Kingdom, and Germany, where there are laws prohibiting it as being promoting violence or social disorder. Social media services such as Facebook and Twitter have been criticized for not having done enough to prohibit the use of their services for attacking people belonging to some specific race, minority etc. BIBREF0 . They have announced though that they would seek to battle against racism and xenophobia BIBREF1 . Nevertheless, the current solutions deployed by them have attempted to address the problem with manual effort, relying on users to report offensive comments BIBREF2 . This not only requires a huge effort by human annotators, but it also has the risk of applying discrimination under subjective judgment. Moreover, a non-automated task by human annotators would have strong impact on system response times, since a computer-based solution can accomplish this task much faster than humans. The massive rise in the user-generated content in the above social media services, with manual filtering not being scalable, highlights the need for automating the process of on-line hate-speech detection.
Despite the fact that the majority of the solutions for automated detection of offensive text rely on Natural Language Processing (NLP) approaches, there is lately a tendency towards employing pure machine learning techniques like neural networks for that task. NLP approaches have the drawback of being complex, and to a large extent dependent on the language used in the text. This provides a strong motivation for employing alternative machine learning models for the classification task. Moreover, the majority of the existing automated approaches depend on using pre-trained vectors (e.g. Glove, Word2Vec) as word embeddings to achieve good performance from the classification model. That makes the detection of hatred content unfeasible in cases where users have deliberately obfuscated their offensive terms with short slang words.
There is a plethora of unsupervised learning models in the existing literature to deal with hate-speech BIBREF3 , as well as in detecting the sentiment polarity in tweets BIBREF4 . At the same time, the supervised learning approaches have not been explored adequately so far. While the task of sentence classification seems similar to that of sentiment analysis; nevertheless, in hate-speech even negative sentiment could still provide useful insight. Our intuition is that the task of hate-speech detection can be further benefited by the incorporation of other sources of information to be used as features into a supervised learning model. A simple statistical analysis on an existing annotated dataset of tweets by BIBREF5 , can easily reveal the existence of significant correlation between the user tendency in expressing opinions that belong to some offensive class (Racism or Sexism), and the annotation labels associated with that class. More precisely, the correlation coefficient value that describes such user tendency was found to be 0.71 for racism in the above dataset, while that value reached as high as 0.76 for sexism. In our opinion, utilizing such user-oriented behavioural data for reinforcing an existing solution is feasible, because such information is retrieva2ble in real-world use-case scenarios like Twitter. This highlights the need to explore the user features more systematically to further improve the classification accuracy of a supervised learning system.
Our approach employs a neural network solution composed of multiple Long-Short-Term-Memory (LSTM) based classifiers, and utilizes user behavioral characteristics such as the tendency towards racism or sexism to boost performance. Although our technique is not necessarily revolutionary in terms of the deep learning models used, we show in this paper that it is quite effective.
Our main contributions are: INLINEFORM0 ) a deep learning architecture for text classification in terms of hateful content, which incorporates features derived form the users' behavioural data, INLINEFORM1 ) a language agnostic solution, due to no-use of pre-trained word embeddings, for detecting hate-speech, INLINEFORM2 ) an experimental evaluation of the model on a Twitter dataset, demonstrating the top performance achieved on the classification task. Special focus is given to investigating how the additional features concerning the users' tendency to utter hate-speech, as expressed by their previous history, could leverage the performance. To the best of our knowledge, there has not been done any previous study on exploring features related to the users tendency in hatred content that used a deep learning model.
The rest of the paper is organized as follows. In Section SECREF2 we describe the problem of hate speech in more detail, and we refer to the existing work in the field in Section SECREF3 . In Section SECREF4 we present our proposed model, while in Section SECREF5 we refer to the dataset used, the evaluation tests we performed and we discuss the results received. Finally, in Section SECREF6 we summarize our contributions and discuss the future work.
Our approach employs a neural network solution composed of multiple Long-Short-Term-Memory (LSTM) based classifiers, and utilizes user behavioral characteristics such as the tendency towards racism or sexism to boost performance.
|
What is used to boost performance?
|
User behavioral characteristics.
|
null | false
| 86
|
The proposed LID algorithm builds on the work in BIBREF8 and BIBREF26. We apply a naive Bayesian classifier with character (2, 4 & 6)-grams, word unigram and word bigram features with a hierarchical lexicon based classifier.
The naive Bayesian classifier is trained to predict the specific language label of a piece of text, but used to first classify text as belonging to either the Nguni family, the Sotho family, English, Afrikaans, Xitsonga or Tshivenda. The scikit-learn multinomial naive Bayes classifier is used for the implementation with an alpha smoothing value of 0.01 and hashed text features.
The lexicon based classifier is then used to predict the specific language within a language group. For the South African languages this is done for the Nguni and Sotho groups. If the lexicon prediction of the specific language has high confidence then its result is used as the final label else the naive Bayesian classifier's specific language prediction is used as the final result. The lexicon is built over all the data and therefore includes the vocabulary from both the training and testing sets.
The lexicon based classifier is designed to trade higher precision for lower recall. The proposed implementation is considered confident if the number of words from the winning language is at least one more than the number of words considered to be from the language scored in second place.
The stacked classifier is tested against three public LID implementations BIBREF17, BIBREF23, BIBREF8. The LID implementation described in BIBREF17 is available on GitHub and is trained and tested according to a post on the fasttext blog. Character (5-6)-gram features with 16 dimensional vectors worked the best. The implementation discussed in BIBREF23 is available from https://github.com/tomkocmi/LanideNN. Following the instructions for an OSX pip install of an old r0.8 release of TensorFlow, the LanideNN code could be executed in Python 3.7.4. Settings were left at their defaults and a learning rate of 0.001 was used followed by a refinement with learning rate of 0.0001. Only one code modification was applied to return the results from a method that previously just printed to screen. The LID algorithm described in BIBREF8 is also available on GitHub.
The stacked classifier is also tested against the results reported for four other algorithms BIBREF16, BIBREF26, BIBREF24, BIBREF15. All the comparisons are done using the NCHLT BIBREF7, DSL 2015 BIBREF19 and DSL 2017 BIBREF1 datasets discussed in Section SECREF2.
We apply a naive Bayesian classifier with character (2, 4 & 6)-grams, word unigram and word bigram features with a hierarchical lexicon based classifier.
|
What is the nature of the classifier？
|
It is a naive Bayesian classifier with character (2, 4 & 6)-grams, word unigram and word bigram features with a hierarchical lexicon based classifier.
|
2003.04032
| false
| null |
We measured intra-annotator agreement between two annotators in three aspects: relations, senses, arguments. To be specific, the annotators’ consistency in annotating the type of a specific relation or sense and the position and scope of arguments are measured. To assess the consistency of annotations and also eliminate coincidental annotations, we used agreement rates, which is calculated by dividing the number of senses under each category where the annotators annotate consistently by the total number of each kind of sense. And considering the potential impact of unbalanced distribution of senses, we also used the Kappa value. And the final agreement study was carried out for the first 300 relations in our corpus. We obtained high agreement results and Kappa value for the discourse relation type and top-level senses ($\ge {0.9} $ ). However, what we did was more than this, and we also achieved great results on the second-level and third-level senses for the sake of our self-demand for high-quality, finally achieving agreement of 0.85 and Kappa value of 0.83 for these two deeper levels of senses.
However, what we did was more than this, and we also achieved great results on the second-level and third-level senses for the sake of our self-demand for high-quality, finally achieving agreement of 0.85 and Kappa value of 0.83 for these two deeper levels of senses.
|
How high is the inter-annotator agreement?
|
The answers are shown as follows:
* agreement of 0.85 and Kappa value of 0.83
|
null | false
| null |
Who won an Oscar for the African Queen
|
Bogart
|
|
null | false
| null |
Give me some ideas for party games at a child's birthday.
|
Some classic children's party games are Pass the Parcel, Musical Chairs, Pin the Tail on the Donkey, Sack Race, Treasure Hunt.
|
|
1909.02480
| false
| null |
Table TABREF40 illustrates the BLEU scores of FlowSeq and baselines with advanced decoding methods such as iterative refinement, IWD and NPD rescoring. The first block in Table TABREF40 includes the baseline results from autoregressive Transformer. For the sampling procedure in IWD and NPD, we sampled from a reduced-temperature model BIBREF11 to obtain high-quality samples. We vary the temperature within $\lbrace 0.1, 0.2, 0.3, 0.4, 0.5, 1.0\rbrace $ and select the best temperature based on the performance on development sets. The analysis of the impact of sampling temperature and other hyper-parameters on samples is in § SECREF50. For FlowSeq, NPD obtains better results than IWD, showing that FlowSeq still falls behind auto-regressive Transformer on model data distributions. Comparing with CMLM BIBREF8 with 10 iterations of refinement, which is a contemporaneous work that achieves state-of-the-art translation performance, FlowSeq obtains competitive performance on both WMT2014 and WMT2016 corpora, with only slight degradation in translation quality. Leveraging iterative refinement to further improve the performance of FlowSeq has been left to future work.
FLOAT SELECTED: Table 2: BLEU scores on two WMT datasets of models using advanced decoding methods. The first block are Transformer-base (Vaswani et al., 2017). The second and the third block are results of models trained w/w.o. knowledge distillation, respectively. n = l × r is the total number of candidates for rescoring.
Table TABREF40 illustrates the BLEU scores of FlowSeq and baselines with advanced decoding methods such as iterative refinement, IWD and NPD rescoring.
FLOAT SELECTED: Table 2: BLEU scores on two WMT datasets of models using advanced decoding methods. The first block are Transformer-base (Vaswani et al., 2017). The second and the third block are results of models trained w/w.o. knowledge distillation, respectively. n = l × r is the total number of candidates for rescoring.
|
What is the performance difference between proposed method and state-of-the-arts on these datasets?
|
Difference is around 1 BLEU score lower on average than state of the art methods.
|
null | false
| null |
Is it better to buy a used car compared to a new car?
|
When it comes to buying a car, there are two main options: buying a new car or buying a used car. While both options have their advantages and disadvantages, many people wonder if it is better to buy a used car compared to a new car. In this article, we will explore the pros and cons of buying a used car and help you decide if it is the right choice for you.
Pros of Buying a Used Car
1. Lower Price
One of the biggest advantages of buying a used car is the lower price. Used cars are typically much cheaper than new cars, even if they are only a few years old. This is because new cars lose value quickly as soon as they are driven off the dealership lot. By buying a used car, you can often get a great deal and save thousands of dollars.
2. Lower Depreciation
New cars depreciate quickly in the first few years of ownership, often losing up to 50% of their value. Used cars, on the other hand, have already gone through this initial depreciation period, so they lose value at a much slower rate. This means that if you decide to sell your used car in a few years, you may not lose as much money as you would with a new car.
3. Lower Insurance Costs
Insurance costs for a used car are typically lower than for a new car. This is because used cars are less expensive, so the insurance company does not have to pay out as much if the car is damaged or totaled. Additionally, if you own the car outright, you may be able to choose a lower level of insurance coverage, which can further reduce your insurance costs.
4. More Options
When you buy a used car, you have a wider range of options to choose from than if you were buying a new car. You can choose from different makes, models, and years, and you may be able to find a car with features that are no longer available on new cars. This means that you can find a car that fits your specific needs and preferences.
5. No Hidden Fees
When you buy a new car, you may be charged various fees and expenses, such as a destination fee, dealer preparation fee, and advertising fee. With a used car, you do not have to worry about these hidden fees, as they have already been paid by the previous owner.
Cons of Buying a Used Car
1. Higher Maintenance Costs
Used cars may require more maintenance and repairs than new cars, especially if they are older or have high mileage. This can result in higher maintenance costs, which can offset some of the savings you get from buying a used car.
2. Higher Interest Rates
If you are financing a used car, you may have to pay a higher interest rate than if you were financing a new car. This is because lenders consider used cars to be a greater risk, as they are more likely to need repairs and maintenance.
3. Limited Warranty
Many used cars do not come with a warranty, or the warranty may have expired. This means that if something goes wrong with the car, you will have to pay for the repairs out of pocket. Some dealerships offer extended warranties for used cars, but these can be expensive.
4. Unknown History
When you buy a used car, you may not know the full history of the vehicle. The previous owner may not have disclosed all accidents or repairs, and there may be hidden issues that only become apparent after you have purchased the car.
5. Less Advanced Features
Used cars may not have the latest technology and safety features that are available on new cars. This can be a disadvantage if you are looking for a car with advanced features, such as a backup camera or lane departure warning.
Conclusion
Buying a used car can be a great option for many people, as it offers a lower price, slower depreciation, lower insurance costs, more options, and no hidden fees. However, there are also some disadvantages to buying a used car, such as higher maintenance costs, limited warranty, unknown history, higher interest rates, and less advanced features. Ultimately, the decision to buy a used car or a new car depends on your personal preferences and budget.
|
|
null | false
| null |
Different ways in which you can eat noodles
|
hand, fork, spoon, chopsticks
|
|
null | false
| null |
Identify which instrument is string or woodwind: Ahenk, Arghul
|
Arghul is woodwind, Ahenk is string.
|
|
null | false
| null |
What are the FAANG companies?
|
Five of the best performing tech stocks of the last decade are Facebook, Amazon, Apple, Netflix, and Google. Those five stocks have been pulled together into an the acronym FAANG.
|
|
null | false
| null |
Which city are Stroopwafels from?
|
Stroopwafels are a type of Dutch waffle-like cake made with ground or shredded wheat and a variety of fillings. They are most commonly found in the city of Haarlem, Netherlands.
|
|
null | false
| 504
|
We first conduct experiments on the MNIST dataset using the LeNet-300-100 and LeNet-5 architectures. In Figure (left), the distribution of weights when training with LLR representation is compared with the case of softmax normalization. Using LLR representation yields a better distribution of weights -the probability of small weights around zero is higher and these weights can be pruned with less impact on the performance. Furthermore, weight decaying can push the weights aggressively towards zero as shown in Figure (right). While softmax normalization primarily affects output layers, the ReLU function may cause overfitting in all layers. Therefore, the effect of weight decaying is more prominent and effective as shown in Figure. This observation is consistent with the analysis and hypotheses from Section 2. Figure shows the trade-off curves for test errors vs. total number of effective parameters for all experiment results on the MNIST dataset. Each curve represents 20 trained networks with the same training settings, each point represents the average total number of weights and average top-1 errors of the pruned networks for each of the 10 different pruning settings. We can see that using LLR representation instead of softmax normalization can reduce the total number of parameters for the same accuracy requirement. Using weight decaying also significantly improve the efficiency of the trained networks. Using both methods yields the most efficient neural networks with better performance than the ones using the iterative pruning approach from, as shown in We found that the AdamW optimizer with weight decaying may increase training accuracy by increasing overfitting and yield less efficient networks, as demonstrated in Figure. Compared with previous results, the optimal pruned model sizes are dramatically increased and using weight decaying does not improve the efficiency of trained networks. Therefore, the trade-off curves can be used to judge the efficiency of trained networks. Curves closer to the bottom-left region on the figure represent more efficient networks. Both ResNet and DenseNet architectures show similar trends in terms of efficiency. The trade-off curves for the same architecture with different initial model sizes seem to be bounded by a single theoretical curve. In the energy efficient region with small number of parameters, the error rate goes down rapidly with a small increase in the number of parameters; while in the high accuracy region, small increases in accuracy require exponentially increasing numbers of parameters. In this regard, the ResNet-32 and the DenseNet-60 architectures may offer better alternative trade-offs in efficiency. For all experiments using either softmax normalization or LLR representation, the same weight-decaying settings are used. In terms of efficiency trade-offs, we see a similar trend as before: linear increase in accuracy tends to require exponential increase in network capacity. We notice that the difference between softmax normalization and LLR representation is less prominent for the DenseNet architecture. One possible reason is that the BSP loss function is not yet fully optimized for the DenseNet architecture. Another possible reason is that the effects of weight decaying are more prominent than softmax normalization for the DenseNet architecture with larger initial model sizes.
We found that the AdamW optimizer with weight decaying may increase training accuracy by increasing overfitting and yield less efficient networks, as demonstrated in Figure 3. Compared with previous results, the optimal pruned model sizes are dramatically increased and using weight decaying does not improve the efficiency of trained networks.****For DenseNet-60 with less than 90K parameters, the performance is comparable to ResNet-56 with 200K parameters. Therefore, overfitting issues with the DenseNet architecture are less prominent than the ResNet architecture. Figure 7 summarize the efficiency trade-offs for both ResNet and DenseNet architectures. Compared with ResNet architecture, the initial DenseNet model sizes are larger, the effects of weight decaying are more prominent than the softmax normalization, and the difference between softmax normalization and LLR representation for DenseNet is smaller.
|
Perhaps providing the evidence of Loshchilov & Hutter (2018)'s AdamW algorithm increasing overparameterization would strengthen the anecdote at the end of S2.3?
|
Additional results and analysis for using AdamW algorithm are added to Section 3.2 to show that the AdamW algorithm may product less efficient networks. Additional analysis of the overfitting issue is also provided in the Appendix.
|
null | false
| 16
|
Data annotation is a major bottleneck for the application of supervised learning approaches to many problems. As a result, unsupervised methods that learn directly from unlabeled data are increasingly important. For tasks related to unsupervised syntactic analysis, discrete generative models have dominated in recent years – for example, for both part-of-speech (POS) induction BIBREF0 , BIBREF1 and unsupervised dependency parsing BIBREF2 , BIBREF3 , BIBREF4 . While similar models have had success on a range of unsupervised tasks, they have mostly ignored the apparent utility of continuous word representations evident from supervised NLP applications BIBREF5 , BIBREF6 . In this work, we focus on leveraging and explicitly representing continuous word embeddings within unsupervised models of syntactic structure.
Pre-trained word embeddings from massive unlabeled corpora offer a compact way of injecting a prior notion of word similarity into models that would otherwise treat words as discrete, isolated categories. However, the specific properties of language captured by any particular embedding scheme can be difficult to control, and, further, may not be ideally suited to the task at hand. For example, pre-trained skip-gram embeddings BIBREF7 with small context window size are found to capture the syntactic properties of language well BIBREF8 , BIBREF9 . However, if our goal is to separate syntactic categories, this embedding space is not ideal – POS categories correspond to overlapping interspersed regions in the embedding space, evident in Figure SECREF4 .
In our approach, we propose to learn a new latent embedding space as a projection of pre-trained embeddings (depicted in Figure SECREF5 ), while jointly learning latent syntactic structure – for example, POS categories or syntactic dependencies. To this end, we introduce a new generative model (shown in Figure FIGREF6 ) that first generates a latent syntactic representation (e.g. a dependency parse) from a discrete structured prior (which we also call the “syntax model”), then, conditioned on this representation, generates a sequence of latent embedding random variables corresponding to each word, and finally produces the observed (pre-trained) word embeddings by projecting these latent vectors through a parameterized non-linear function. The latent embeddings can be jointly learned with the structured syntax model in a completely unsupervised fashion.
By choosing an invertible neural network as our non-linear projector, and then parameterizing our model in terms of the projection's inverse, we are able to derive tractable exact inference and marginal likelihood computation procedures so long as inference is tractable in the underlying syntax model. In sec:learn-with-inv we show that this derivation corresponds to an alternate view of our approach whereby we jointly learn a mapping of observed word embeddings to a new embedding space that is more suitable for the syntax model, but include an additional Jacobian regularization term to prevent information loss.
Recent work has sought to take advantage of word embeddings in unsupervised generative models with alternate approaches BIBREF9 , BIBREF10 , BIBREF11 , BIBREF12 . BIBREF9 build an HMM with Gaussian emissions on observed word embeddings, but they do not attempt to learn new embeddings. BIBREF10 , BIBREF11 , and BIBREF12 extend HMM or dependency model with valence (DMV) BIBREF2 with multinomials that use word (or tag) embeddings in their parameterization. However, they do not represent the embeddings as latent variables.
In experiments, we instantiate our approach using both a Markov-structured syntax model and a tree-structured syntax model – specifically, the DMV. We evaluate on two tasks: part-of-speech (POS) induction and unsupervised dependency parsing without gold POS tags. Experimental results on the Penn Treebank BIBREF13 demonstrate that our approach improves the basic HMM and DMV by a large margin, leading to the state-of-the-art results on POS induction, and state-of-the-art results on unsupervised dependency parsing in the difficult training scenario where neither gold POS annotation nor punctuation-based constraints are available.
By choosing an invertible neural network as our non-linear projector, and then parameterizing our model in terms of the projection’s inverse, we are able to derive tractable exact inference and marginal likelihood computation procedures so long as inference is tractable in the underlying syntax model.
|
In their approach, what is the only condition that the authors can derive tractable exact inference and marginal likelihood computation procedures?
|
Tractable in the underlying syntax model.
|
null | false
| null |
Why are people scared of snakes?
|
Snakes tend to scare people for a number of reasons. For starters, many of them have poisonous venom. Snakes are also very stealthy and oftentimes you may not even be aware of their presence. Some people are also frightened by the fact that snakes slither, and do not have any legs. Encountering a snake in the wild can send chills down your spine.
|
|
null | false
| null |
What does a 9 carat gold mean
|
Pure gold with no additives is knows as 24 carat gold. An item marked as 9 carat gold contains 9/24th (or 37.5 percent) of pure gold and the rest are additives, typically a mix of silver, copper, zinc, etc.
|
|
null | false
| null |
What is the best soccer club in the premier league?
|
The answer to this question is subjective. Many believe that Manchester United, a soccer club founded in 1878, is the best soccer club from the premier league. The club has won 20 first division titles and is a 3 time champion league winner. However there are also many other opinions. For example, a lots of fans consider that Liverpool, which holds the record for the most champion league title wins from premier league, is the best club in the league. Manchester City has also been performing great in the past decade and has been considered the best club in the league for the past 10 years.
|
|
null | false
| 111
|
We utilised two datasets for both Turkish and English to evaluate our methods.
For Turkish, as the first dataset, we utilised the movie reviews which are collected from a popular website. The number of reviews in this movie corpus is 20,244 and the average number of words in reviews is 39. Each of these reviews has a star-rating score which is indicative of sentiment. These polarity scores are between the values 0.5 and 5, at intervals of 0.5. We consider a review to be negative it the score is equal to or lower than 2.5. On the other hand, if it is equal to or higher than 4, it is assumed to be positive. We have randomly selected 7,020 negative and 7,020 positive reviews and processed only them.
The second Turkish dataset is the Twitter corpus which is formed of tweets about Turkish mobile network operators. Those tweets are mostly much noisier and shorter compared to the reviews in the movie corpus. In total, there are 1,716 tweets. 973 of them are negative and 743 of them are positive. These tweets are manually annotated by two humans, where the labels are either positive or negative. We measured the Cohen's Kappa inter-annotator agreement score to be 0.82. If there was a disagreement on the polarity of a tweet, we removed it.
We also utilised two other datasets in English to test the cross-linguality of our approaches. One of them is a movie corpus collected from the web. There are 5,331 positive reviews and 5,331 negative reviews in this corpus. The other is a Twitter dataset, which has nearly 1.6 million tweets annotated through a distant supervised method BIBREF14. These tweets have positive, neutral, and negative labels. We have selected 7,020 positive tweets and 7,020 negative tweets randomly to generate a balanced dataset.
We measured the Cohen's Kappa inter-annotator agreement score to be 0.82.
|
How about the inter-annotator agreement status in the Turkish tweets corpus？
|
They measured the Cohen's Kappa inter-annotator agreement score to be 0.82.
|
null | false
| null |
What are the causes Alzheimer's disease?
|
Undrestanding the causes of the Alzheimer's disease is an ongoing research. The research so far indicate that causes probably include a combination of age-related changes in the brain, along with genetic, environmental, and lifestyle factors. The importance of any one of these factors may differ from person to person.
One well-known gene that influences Alzheimer’s risk is the apolipoprotein E (APOE) gene, which has been identified via Genome Wide Association Studies (GWAS). The APOE gene is involved in making a protein that helps carry cholesterol and other types of fat in the bloodstream.
|
|
null | false
| 329
|
With the growth of the internet in the last decades, users are faced with an increasing amount of information and have to find ways to summarize it. However, producing summaries in a multi-document setting is a challenging task; the language used to display the same information in a sentence can vary significantly, making it difficult for summarization models to capture. Thus large corpora are needed to develop efficient models. There exist two types of summarization: extractive and abstractive. Extractive summarization outputs summaries in two steps, namely via sentence ranking, where an importance score is assigned to each sentence, and via the subsequent sentence selection, where the most appropriate sentence is chosen. In abstractive summarization, summaries are generated word by word auto-regressively, using sequence-to-sequence or language models. Given the complexity of multi-document summarization and the lack of datasets, most researchers use extractive summarization and rely on hand-crafted features or additional annotated data, both needing human expertise.
To our knowledge, wiki2018 is the only work that has proposed a large dataset for multi-document summarization. By considering Wikipedia entries as a collection of summaries on various topics given by their title (e.g., Machine Learning, Stephen King), they create a dataset of significant size, where the lead section of an article is defined as the reference summary and input documents are a mixture of pages obtained from the article's reference section and a search engine. While this approach benefits from the large number of Wikipedia articles, in many cases, articles contain only a few references that tend to be of the desired high quality, and most input documents end up being obtained via a search engine, which results in noisy data. Moreover, at testing time no references are provided, as they have to be provided by human contributors. wiki2018 showed that in this case, generated summaries based on search engine results alone are of poor quality and cannot be used.
In contrast, we propose a novel domain-specific dataset containing $14\,652$ samples, based on professional video game reviews obtained via Metacritic and gameplay sections from Wikipedia. By using Metacritic reviews in addition to Wikipedia articles, we benefit from a number of factors. First, the set of aspects used to assess a game is limited and consequently, reviews share redundancy. Second, because they are written by professional journalists, reviews tend to be in-depth and of high-quality. Additionally, when a video game is released, journalists have an incentive to write a complete review and publish it online as soon as possible to draw the attention of potential customers and increase the revenue of their website BIBREF0. Therefore, several reviews for the same product become quickly available and the first version of the corresponding Wikipedia page is usually made available shortly after. Lastly, reviews and Wikipedia pages are available in multiple languages, which opens up the possibility for multilingual multi-document summarization.
In contrast, we propose a novel domain-specific dataset containing $14\,652$ samples, based on professional video game reviews obtained via Metacritic and gameplay sections from Wikipedia.
|
What dataset does the paper propose?
|
A novel domain-specific dataset.
|
null | false
| null |
The 98th General Assembly enacted a total of 1,175 bills into law. Notable among these was the Illinois Religious Freedom and Marriage Fairness Act, which removed legal barriers to same-sex marriage in Illinois. The House of Representatives passed the bill by a 61–54 vote on November 5, 2013, and the Senate passed it by a 32–21 vote. The governor signed the bill into law on November 20, and the law took effect on June 1, 2014.
The General Assembly also passed the Illinois Bill of Rights for the Homeless, which made Illinois the second state to ban discrimination against the homeless.
The General Assembly also passed the Firearm Concealed Carry Act (FCCA), a change to Illinois gun law that made Illinois the last state in the country to enact concealed carry. Passed over the governor's veto, the FCCA took effect "mere hours before the seventh circuit's deadline" in Moore v. Madigan.
Late in the legislative session, the General Assembly passed a bill that changes the original Illinois wiretapping law adding that in order to commit a criminal offense, a person must be recording "in a surreptitious manner". On December 30, 2014, Governor Quinn signed the bill into law as Public Act 098-1142. The bill's sponsors, Elaine Nekritz and Kwame Raoul, claimed the law upholds the rights of citizens to record in public.
The General Assembly approved five ballot measures for the 2014 Illinois general election, an unusually large number for the state, including three advisory referendums and two constitutional amendments. The constitutional amendments, Marsy's Law and the Illinois Right to Vote Amendment, both passed with the required 60% majority of the popular vote.
|
here is a blurb about the 98th Illinois General Assembly, what were the notable laws that were passed?
|
removing the legal barriers to same-sex marriage, banning discrimination against homeless people, conceal and carry firearms, and revising the wiretapping laws so that citizens can record in public.
|
null | false
| null |
What basic tools should I keep around the house?
|
Keeping tools around the house to make simple fixes is super important as a homeowner and a renter. Furthermore, being self sufficient and not relying on others to do basic work is very empowering. Some basic tools that everyone should keep around the house are a screwdriver with various different heads (eg. phillips, flat head, allen wrench, etc), needle nose pliers, channel lock pliers, slip joint pliers an electric drill, drill bit set, and a hammer. You can accomplish a lot of small fixes and have on hand when that emergency situation arises.
|
|
null | false
| null |
What teams did Tom Brady play for when he won the Super Bowl?
|
Tom Brady won the Super Bowl while playing for the New England Patriots and Tampa Bay Buccaneers.
|
|
1912.10435
| false
| null |
We first focused on directed coattention via context to query and query to context attention as discussed in BIDAF BIBREF9. We then implemented localized feature extraction by 1D convolutions to add local information to coattention based on the QANET architecture BIBREF10. Subsequently, we experimented with different types of skip connections to inject BERT embedding information back into our modified network. We then applied what we learned using the base BERT model to the large BERT model. Finally, we performed hyperparameter tuning by adjusting the number of coattention blocks, the batch size, and the number of epochs trained and ensembled our three best networks. Each part of the project is discussed further in the subsections below.
Finally, we performed hyperparameter tuning by adjusting the number of coattention blocks, the batch size, and the number of epochs trained and ensembled our three best networks.
|
What hyperparameters have been tuned?
|
The answers are shown as follows:
* number of coattention blocks, the batch size, and the number of epochs trained and ensembled our three best networks
|
null | false
| null |
Which is a species of fish? Hoki or Loki
|
Hoki
|
|
null | false
| null |
What is Christmas?
|
Christmas is a Christian holiday that represents when Jesus was born. This holiday is celebrated on December 25 every year. Church's will have a manger setup with figures showing what the birth of Christ would have looked like thousands of years ago. The figures represented would be the wise men, mary, joseph, and Jesus in the manger. The wise men were led to the manger of Jesus by following a star in the sky. Billions of people all over the world celebrate this holiday every year.
|
|
1708.00077
| false
| null |
We use the dropout technique of Gal & Ghahramani gal as a baseline because it is the most similar dropout technique to our approach and denote it VBD (variational binary dropout).
We use the dropout technique of Gal & Ghahramani gal as a baseline because it is the most similar dropout technique to our approach and denote it VBD (variational binary dropout).
|
What is binary variational dropout?
|
The answers are shown as follows:
* the dropout technique of Gal & Ghahramani gal
|
null | false
| null |
List the four season
|
Winter, Spring, Summer Fall
|
|
null | false
| null |
In West Africa, the decline of the Atlantic slave trade in the 1820s caused dramatic economic shifts in local polities. The gradual decline of slave-trading, prompted by a lack of demand for slaves in the New World, increasing anti-slavery legislation in Europe and America, and the British Royal Navy's increasing presence off the West African coast, obliged African states to adopt new economies. Between 1808 and 1860, the British West Africa Squadron seized approximately 1,600 slave ships and freed 150,000 Africans who were aboard.
Action was also taken against African leaders who refused to agree to British treaties to outlaw the trade, for example against "the usurping King of Lagos", deposed in 1851. Anti-slavery treaties were signed with over 50 African rulers. The largest powers of West Africa (the Asante Confederacy, the Kingdom of Dahomey, and the Oyo Empire) adopted different ways of adapting to the shift. Asante and Dahomey concentrated on the development of "legitimate commerce" in the form of palm oil, cocoa, timber and gold, forming the bedrock of West Africa's modern export trade. The Oyo Empire, unable to adapt, collapsed into civil wars.
|
What new economies developed because of the anti-slave treaties?
|
Instead of exporting slaves, Africa began exporting palm oil, cocoa, timber, and gold.
|
null | false
| 252
|
SVM A Support Vector Machines model in BIBREF36 detects misinformation relying on manually extracted features.
CNN A Convolutional Neural Network model BIBREF37 employs pre-trained word embeddings based on Word2Vec as input embeddings to capture features similar to n-grams.
TE Tensor Embeddings BIBREF38 leverages tensor decomposition to derive concise claim embeddings, which are used to create a claim-by-claim graph for label propagation.
DeClarE Evidence-Aware Deep Learning BIBREF39 encodes claims and articles by Bi-LSTM and focuses on each other based on attention mechanism, and then concatenates claim source and article source information.
MTL-LSTM A multi-task learning model based on LSTM networks BIBREF14 trains jointly the tasks of veracity classification, rumor detection, and stance detection.
TRNN Tree-structured RNN BIBREF40 is a bottom-up and a top-down tree-structured model based on recursive neural networks.
Bayesian-DL Bayesian Deep Learning model BIBREF41 first adopts Bayesian to represent both the prediction and uncertainty of claim and then encodes replies based on LSTM to update and generate a posterior representations.
SVM A Support Vector Machines model in (Derczynski et al., 2017) detects misinformation relying on manually extracted features. CNN A Convolutional Neural Network model (Chen et al., 2017) employs pre-trained word embeddings based on Word2Vec as input embeddings to capture features similar to n-grams. TE Tensor Embeddings (Guacho et al., 2018) leverages tensor decomposition to derive concise claim embeddings, which are used to create a claimby-claim graph for label propagation. DeClarE Evidence-Aware Deep Learning (Popat et al., 2018) encodes claims and articles by Bi-LSTM and focuses on each other based on attention mechanism, and then concatenates claim source and article source information. MTL-LSTM A multi-task learning model based on LSTM networks (Kochkina et al., 2018) trains jointly the tasks of veracity classification, rumor detection, and stance detection. TRNN Tree-structured RNN (Ma et al., 2018b) is a bottom-up and a top-down tree-structured model based on recursive neural networks. Bayesian-DL Bayesian Deep Learning model (Zhang et al., 2019) first adopts Bayesian to represent both the prediction and uncertainty of claim and then encodes replies based on LSTM to update and generate a posterior representations.
|
What baselines do they use in the paper?
|
SVM, CNN, TE, DeClarE, MTL-LSTM, TRNN, and Bayesian-DL.
|
null | false
| 35
|
The word embeddings of the source and target language sentences are used as initial vectors of the model to improve contextualization. The skip gram model of the word2vec algorithm optimizes the vectors by accounting for the average log probability of context words given a source word. DISPLAYFORM0
where k is the context window taken for the vectorization, INLINEFORM0 refers to the INLINEFORM1 word of the corpus and INLINEFORM2 is the size of the training corpus in terms of the number of words. Here, the probabily INLINEFORM3 is computed as a hierarchical softmax of the product of the transpose of the output vector of INLINEFORM4 and the input vector of INLINEFORM5 for each and every pair over the entire vocabulary. The processes of negative sampling and subsampling of frequent words that were used in the original model aren't used in this experiment BIBREF3 .
For the process of creating semantically meaningful word embeddings, a monolingual corpus of 569,772 Tamil sentences was used. This gave the vectors more contextual richness due to the increased size of the corpus as opposed to using just the bilingual corpus' target side sentences BIBREF3 .
In the experiment, the word2vec model was trained using a vector size of 100 to ensure that the bulk of the limited memory of the GPU will be used for the neural attention translation model. It has been shown that any size over that of 150 used for word vectorization gives similar results and that a size of 100 performs close to the model with 150-sized word vectors BIBREF7 . A standard size of 5 was used as window size and the model was trained over 7 worker threads simultaneously. A batch size of 50 words was used for training. The negative sampling was set at 1 as it is the nature of morphologically rich languages to have a lot of important words that don't occur more than once in the corpus. The gensim word2vec toolkit was used to implement this word embedding process BIBREF8 .
The processes of negative sampling and subsampling of frequent words that were used in the original model aren't used in this experiment (Mikolov and Dean 2013).
|
Is negative sampling used in the experiment？
|
No.
|
null | false
| 16
|
Data annotation is a major bottleneck for the application of supervised learning approaches to many problems. As a result, unsupervised methods that learn directly from unlabeled data are increasingly important. For tasks related to unsupervised syntactic analysis, discrete generative models have dominated in recent years – for example, for both part-of-speech (POS) induction BIBREF0 , BIBREF1 and unsupervised dependency parsing BIBREF2 , BIBREF3 , BIBREF4 . While similar models have had success on a range of unsupervised tasks, they have mostly ignored the apparent utility of continuous word representations evident from supervised NLP applications BIBREF5 , BIBREF6 . In this work, we focus on leveraging and explicitly representing continuous word embeddings within unsupervised models of syntactic structure.
Pre-trained word embeddings from massive unlabeled corpora offer a compact way of injecting a prior notion of word similarity into models that would otherwise treat words as discrete, isolated categories. However, the specific properties of language captured by any particular embedding scheme can be difficult to control, and, further, may not be ideally suited to the task at hand. For example, pre-trained skip-gram embeddings BIBREF7 with small context window size are found to capture the syntactic properties of language well BIBREF8 , BIBREF9 . However, if our goal is to separate syntactic categories, this embedding space is not ideal – POS categories correspond to overlapping interspersed regions in the embedding space, evident in Figure SECREF4 .
In our approach, we propose to learn a new latent embedding space as a projection of pre-trained embeddings (depicted in Figure SECREF5 ), while jointly learning latent syntactic structure – for example, POS categories or syntactic dependencies. To this end, we introduce a new generative model (shown in Figure FIGREF6 ) that first generates a latent syntactic representation (e.g. a dependency parse) from a discrete structured prior (which we also call the “syntax model”), then, conditioned on this representation, generates a sequence of latent embedding random variables corresponding to each word, and finally produces the observed (pre-trained) word embeddings by projecting these latent vectors through a parameterized non-linear function. The latent embeddings can be jointly learned with the structured syntax model in a completely unsupervised fashion.
By choosing an invertible neural network as our non-linear projector, and then parameterizing our model in terms of the projection's inverse, we are able to derive tractable exact inference and marginal likelihood computation procedures so long as inference is tractable in the underlying syntax model. In sec:learn-with-inv we show that this derivation corresponds to an alternate view of our approach whereby we jointly learn a mapping of observed word embeddings to a new embedding space that is more suitable for the syntax model, but include an additional Jacobian regularization term to prevent information loss.
Recent work has sought to take advantage of word embeddings in unsupervised generative models with alternate approaches BIBREF9 , BIBREF10 , BIBREF11 , BIBREF12 . BIBREF9 build an HMM with Gaussian emissions on observed word embeddings, but they do not attempt to learn new embeddings. BIBREF10 , BIBREF11 , and BIBREF12 extend HMM or dependency model with valence (DMV) BIBREF2 with multinomials that use word (or tag) embeddings in their parameterization. However, they do not represent the embeddings as latent variables.
In experiments, we instantiate our approach using both a Markov-structured syntax model and a tree-structured syntax model – specifically, the DMV. We evaluate on two tasks: part-of-speech (POS) induction and unsupervised dependency parsing without gold POS tags. Experimental results on the Penn Treebank BIBREF13 demonstrate that our approach improves the basic HMM and DMV by a large margin, leading to the state-of-the-art results on POS induction, and state-of-the-art results on unsupervised dependency parsing in the difficult training scenario where neither gold POS annotation nor punctuation-based constraints are available.
By choosing an invertible neural network as our non-linear projector, and then parameterizing our model in terms of the projection’s inverse, we are able to derive tractable exact inference and marginal likelihood computation procedures so long as inference is tractable in the underlying syntax model.
|
What is chosen as their non-linear projector?
|
An invertible neural network.
|
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.
With this in mind, we organised the shared task on fine-grained propaganda detection at the NLP4IF@EMNLP-IJCNLP 2019 workshop.
|
What kind of task was organised?
|
The shared task on fine-grained propaganda detection.
|
null | false
| null |
Bernd Rosemeyer (14 October 1909 – 28 January 1938) was a german racing driver andd speed record holder. He is considered on of the greatest racing driver of all time. Tough he was no a member of the Nazi party, she was made a member of the SS for propaganda purposes and hold the rank off Hauptsturmführer.
|
Fix the typos and grammar below
|
Bernd Rosemeyer (14 October 1909 – 28 January 1938) was a German racing driver and speed record holder. He is considered one of the greatest racing drivers of all time. Though he was not a member of the Nazi party, he was made a member of the SS for propaganda purposes and held the rank of Hauptsturmführer.
|
null | false
| null |
What is System Thinking
|
Systems thinking is a way of making sense of complexity by looking at it in terms of wholes and relationships rather than splitting it into parts. It is based on the idea that all critical processes in an organization are interrelated and work together to achieve a common goal.
|
|
1912.10806
| false
| null |
The data for this project are two parts, the first part is the historical S&P 500 component stocks, which are downloaded from the Yahoo Finance. We use the data over the period of from 12/07/2017 to 06/01/2018. The second part is the news article from financial domain are collected with the same time period as stock data. Since our paper illustrates the relationship between the sentiment of the news articles and stocks' price. Hence, only news article from financial domain are collected. The data is mainly taken from Webhose archived data, which consists of 306242 news articles present in JSON format, dating from December 2017 up to end of June 2018. The former 85% of the dataset is used as the training data and the remainder 15% is used as the testing data. The News publishers for this data are CNBC.com, Reuters.com, WSJ.com, Fortune.com. The Wall Street Journal is one of the largest newspapers in the United States, which coverage of breaking news and current headlines from the US and around the world include top stories, photos, videos, detailed analysis and in-depth thoughts; CNBC primarily carries business day coverage of U.S. and international financial markets, which following the end of the business day and on non-trading days; Fortune is an American multinational business magazine; Reuters is an international news organization. We preprocess the raw article body and use NLTK sentiment package alence Aware Dictionary and Sentiment Reasoner (VADER) to extract sentiment scores.
The data for this project are two parts, the first part is the historical S&P 500 component stocks, which are downloaded from the Yahoo Finance. We use the data over the period of from 12/07/2017 to 06/01/2018. The second part is the news article from financial domain are collected with the same time period as stock data.
Hence, only news article from financial domain are collected. The data is mainly taken from Webhose archived data, which consists of 306242 news articles present in JSON format, dating from December 2017 up to end of June 2018.
|
What is the dataset used in the paper?
|
historical S&P 500 component stocks
306242 news articles
|
null | false
| null |
Solitary bees, such as leafcutters, do not form colonies. Unlike social insects (ants, yellow jackets, honeybees), leafcutters work alone building isolated nests. Similar to honeybees, female bees perform nearly all essential tasks of brood rearing. These native insects perform essential tasks, pollinating wild plants. The alfalfa leaf cutter bee (Megachile rotundata), native to Europe, has been semi-domesticated for crop pollination. In North America, the species was deliberately imported to assist in the pollination of food crops, but has now become feral and widespread.
|
Are leafcutters solitary or social insects?
|
solitary
|
null | false
| null |
Few creative ways we can market products and services in modern era?
|
1. Influencer Partnership
2. Social Media Campaign
3. Events and experiential marketing
|
|
null | false
| 58
|
The race and gender bias evaluation was carried out on the output of the 219 automatic systems that participated in SemEval-2018 Task 1: Affect in Tweets BIBREF5 . The shared task included five subtasks on inferring the affectual state of a person from their tweet: 1. emotion intensity regression, 2. emotion intensity ordinal classification, 3. valence (sentiment) regression, 4. valence ordinal classification, and 5. emotion classification. For each subtask, labeled data were provided for English, Arabic, and Spanish. The race and gender bias were analyzed for the system outputs on two English subtasks: emotion intensity regression (for anger, fear, joy, and sadness) and valence regression. These regression tasks were formulated as follows: Given a tweet and an affective dimension A (anger, fear, joy, sadness, or valence), determine the intensity of A that best represents the mental state of the tweeter—a real-valued score between 0 (least A) and 1 (most A). Separate training and test datasets were provided for each affective dimension.
Training sets included tweets along with gold intensity scores. Two test sets were provided for each task: 1. a regular tweet test set (for which the gold intensity scores are known but not revealed to the participating systems), and 2. the Equity Evaluation Corpus (for which no gold intensity labels exist). Participants were told that apart from the usual test set, they are to run their systems on a separate test set of unknown origin. The participants were instructed to train their system on the tweets training sets provided, and that they could use any other resources they may find or create. They were to run the same final system on the two test sets. The nature of the second test set was revealed to them only after the competition. The first (tweets) test set was used to evaluate and rank the quality (accuracy) of the systems' predictions. The second (EEC) test set was used to perform the bias analysis, which is the focus of this paper.
Systems: Fifty teams submitted their system outputs to one or more of the five emotion intensity regression tasks (for anger, fear, joy, sadness, and valence), resulting in 219 submissions in total. Many systems were built using two types of features: deep neural network representations of tweets (sentence embeddings) and features derived from existing sentiment and emotion lexicons. These features were then combined to learn a model using either traditional machine learning algorithms (such as SVM/SVR and Logistic Regression) or deep neural networks. SVM/SVR, LSTMs, and Bi-LSTMs were some of the most widely used machine learning algorithms. The sentence embeddings were obtained by training a neural network on the provided training data, a distant supervision corpus (e.g., AIT2018 Distant Supervision Corpus that has tweets with emotion-related query terms), sentiment-labeled tweet corpora (e.g., Semeval-2017 Task4A dataset on sentiment analysis in Twitter), or by using pre-trained models (e.g., DeepMoji BIBREF20 , Skip thoughts BIBREF21 ). The lexicon features were often derived from the NRC emotion and sentiment lexicons BIBREF22 , BIBREF23 , BIBREF24 , AFINN BIBREF25 , and Bing Liu Lexicon BIBREF26 .
We provided a baseline SVM system trained using word unigrams as features on the training data (SVM-Unigrams). This system is also included in the current analysis.
Measuring bias: To examine gender bias, we compared each system's predicted scores on the EEC sentence pairs as follows:
Thus, eleven pairs of scores (ten pairs of scores from ten noun phrase pairs and one pair of scores from the averages on name subsets) were examined for each template–emotion word instantiation. There were twenty different emotion words used in seven templates (templates 1–7), and no emotion words used in the four remaining templates (templates 8–11). In total, INLINEFORM0 pairs of scores were compared.
Similarly, to examine race bias, we compared pairs of system predicted scores as follows:
Thus, one pair of scores was examined for each template–emotion word instantiation. In total, INLINEFORM0 pairs of scores were compared.
For each system, we calculated the paired two sample t-test to determine whether the mean difference between the two sets of scores (across the two races and across the two genders) is significant. We set the significance level to 0.05. However, since we performed 438 assessments (219 submissions evaluated for biases in both gender and race), we applied Bonferroni correction. The null hypothesis that the true mean difference between the paired samples was zero was rejected if the calculated p-value fell below INLINEFORM0 .
To examine gender bias, we compared each system’s predicted scores on the EEC sentence pairs as follows: • We compared the predicted intensity score for a sentence generated from a template using a female noun phrase (e.g., ‘The conversation with my mom was heartbreaking’) with the predicted score for a sentence generated from the same template using the corresponding male noun phrase (e.g., ‘The conversation with my dad was heartbreaking’). • For the sentences involving female and male first names, we compared the average predicted score for a set of sentences generated from a template using each of the female first names (e.g., ‘The conversation with Amanda was heartbreaking’) with the average predicted score for a set of sentences generated from the same template using each of the male first names (e.g., ‘The conversation with Alonzo was heartbreaking’).
|
What did the author do to examine gender bias?
|
• They compared the predicted intensity score for a sentence generated from a template using a female noun phrase with the predicted score for a sentence generated from the same template using the corresponding male noun phrase. • For the sentences involving female and male first names, they compared the average predicted score for a set of sentences generated from a template using each of the female first names with the average predicted score for a set of sentences generated from the same template using each of the male first names.
|
null | false
| null |
How can I improve my sleep?
|
There are a number of variables that can help you improve your sleep. It is important to try different recommendations and see which ones will have the biggest impact for you. Sleep is the foundation of our mental and physical health and performance in all endeavors.
1) View sunlight by going outside within 30-60 minutes of waking. Do that again in the late afternoon, prior to sunset. If you wake up before the sun is out and you want to be awake, turn on artificial lights and then go outside once the sun rises.
On bright cloudless days: view morning and afternoon sun for 10 min; cloudy days: 20 min; very overcast days 30-60 min. If you live someplace with very minimal light, consider an artificial daytime simulator source.
Don’t wear sunglasses for this practice if you safely can, but contact lenses and eyeglasses are fine.
No, you don’t have to look directly at the sun, and never look at ANY light so bright it is painful to view! That said, you can’t wear a brimmed hat, sunglasses and remain in the shade and expect to “wake up” your circadian clock.
2) Wake up at the same time each day and go to sleep when you first start to feel sleepy. Pushing through the sleepy late evening feeling and going to sleep too late is one reason people wake at 3 am and can’t fall back asleep.
3) Avoid caffeine within 8-10 hours of bedtime.
4) Avoid viewing bright lights—especially bright overhead lights between 10 pm and 4 am. Here is a simple rule: only use as much artificial lighting as is necessary for you to remain and move about safely at night. Blue blockers can help a bit at night but still dim the lights. Viewing bright lights of all colors are a problem for your circadian system. Candlelight and moonlight are fine.
5) Limit daytime naps to less than 90 min, or don’t nap at all.
6) If you wake up in the middle of the night (which, by the way, is normal to do once or so each night) but you can’t fall back asleep, consider doing an NSDR protocol when you wake up.
7) Keep the room you sleep in cool and dark and layer on blankets that you can remove.
Your body needs to drop in temperature by 1-3 degrees to fall and stay asleep effectively. Body temperature increases are one reason you wake up. Thus, keep your room cool and remove blankets as needed. If it’s too hot you would have to use a cooling device and that’s harder than simply tossing off blankets if you get too warm.
8) Drinking alcohol messes up your sleep. As do most sleep medications.
9) Kids have changing sleep needs over time. Adjust accordingly.
Teens might be night owls at 15 but become “morning people” as they age or need 6 hours a night in summer and 7-8 in winter. It will vary.
|
|
null | false
| null |
What is the best electrical system for RVs?
|
There have been recent improvements in RV electrical systems. At the core, RVs run off “house batteries”, or large batteries that power the RV portion of the vehicle. There are several types of batteries available, and several ways of powering these batteries. The two main types of batteries are lead acid, and lithium. Lithium batteries are newer technology and will not be found in older RVs unless retrofitted. Lithium batteries have a larger charge capacity and can discharge safely further than lead acid batteries. However, lead acid batteries are cheaper. There are several ways of charging these batteries. The most common is a plug that provides either 30A or 50A of power, depending on the RV size. Most RVs also have generators, but generators are noisy and require either gas or diesel to run. Some RVs with lithium batteries (or more rarely, lead acid), have a second alternator installed that charges the batteries while driving. Finally, solar power is a popular option. Solar power is completely quiet and renewable, but can be expensive to install and is slower to charge batteries. Most RVs have a combination of the above, and consumers choose the system that best fits their lifestyle.
|
|
null | false
| null |
If I give you a list of weapons, can you break them into melee and ranged. The list is knife, bow, axe, shotgun, mace, whip, rifle and cannon.
|
Melee weapons include the knife, axe, mace and whip. Ranged weapons include the bow, shotgun, rifle and cannon.
|
|
null | false
| null |
The Needle is a fictional character, a mutant supervillain appearing in American comic books published by Marvel Comics. Created by Mark Gruenwald, Carmine Infantino, and Al Gordon, the character first appeared in Spider-Woman #9 (December 1978). Creator Mark Gruenwald brought the character back during his run on the West Coast Avengers as a member of the villain team Night Shift, which included other former Spider-Woman foes.
After being mugged one night while leaving work, Josef Saint, an elderly tailor, lost an eye and his ability to speak. While recuperating, he discovered that he has the power to paralyze with his gaze. Seeking revenge, Saint assumed the guise of the vigilante the Needle. Armed with a yard-long needle, the mute villain attacked young men on the streets at night, sewing their mouths shut. After victimizing S.H.I.E.L.D. agent Jerry Hunt, the Needle incurred the wrath of Spider-Woman, the agent's girlfriend, and she took him out with a close-range venom blast. He was subsequently arrested.
With other superhuman adventurers and criminals, the Needle was imprisoned by the Locksmith, but then freed by Spider-Woman.
The Needle joined the Night Shift, and teamed with Captain America against the Power Broker and his augmented mutates. Alongside the Night Shift, he tested Moon Knight to take over as the leader of the Night Shift. Later, alongside the Night Shift, he battled the West Coast Avengers. Alongside the Night Shift, the second Hangman, and Satannish, he battled the Avengers West Coast; his abilities had been enhanced by Satannish's black magic.
The Needle was later defeated by Armory.
Needle appears with the Night Shift, as part of the Hood's gang. They battle the Midnight Sons, and Needle impales Daimon Hellstrom with his needle, before Jennifer Kale knocks him off with a spell. He and the Night Shift are killed when the zombie virus mutates and becomes airborne. The virus cloud begins to rain blood, and reanimates the Night Shift as zombies. Dormammu assumes control of the Night Shift and uses them to fight the Midnight Sons. When Jennifer Kale and the Black Talon contain the virus within the Zombie (Simon Garth), the Night Shift members are apparently restored to normal, and the Hood teleports away with them.
|
Please provide a short biography of The Needle from the passage provided.
|
The Needle is a mutant supervillain created by Mark Gruenwald, Carmine Infantino, and Al Gordon. He first appeared in Spider-Woman #9 (December 1978) and was brought back during his run on the West Coast Avengers as a member of the villain team Night Shift. He was imprisoned by the Locksmith and freed by Spider-Woman. He joined the Night Shift and teamed with Captain America against the Power Broker and his augmented mutates. He also battled the West Coast Avengers, the second Hangman, and Satannish.
He was later defeated by Armory. Needle appears with the Night Shift, as part of the Hood's gang, and they battle the Midnight Sons. They are killed when the zombie virus mutates and becomes airborne. Dormammu assumes control of the Night Shift and uses them to fight the Midnight Sons. When Jennifer Kale and the Black Talon contain the virus within the Zombie, the Night Shift members are restored to normal and the Hood teleports away with them.
|
null | false
| 46
|
We utilize fasttext BIBREF18 to generate the embeddings of all words from both Text-SQuAD and Spoken-SQuAD. We adopt the phoneme sequence embeddings to replace the original character sequence embeddings using the method proposed by Li et al. BIBREF5 . The source domain model and the target domain model share the same set of word embedding matrix to improve the alignment between these two domains.
W-GAN is adopted for our domain discriminator BIBREF19 , which stacks 5 residual blocks of 1D convolutional layers with 96 filters and filter size 5 followed by one linear layer to convert each input vector sequence into one scalar value.
All models used in the experiments are trained with batch size 20, using adam with learning rate INLINEFORM0 and the early stop strategy. The dimension of the hidden state is set to 96 for all layers, and the number of self-attention heads is set to 2. The setup is slightly different but better than the setting suggested by the original QAnet.
We utilize fasttext [19] to generate the embeddings of all words from both Text-SQuAD and Spoken-SQuAD.
|
What's the utilization of fast text?
|
To generate the embeddings of all words from both Text-SQuAD and Spoken-SQuAD
|
1806.04524
| false
| null |
Using our platform, we extracted anonymized user interaction data in the manner of real quizzes generated for a collection of several input video sources. We obtained a corpus of approximately 300,000 sentences, from which roughly 1.5 million single-quiz question training examples were derived. We split this dataset using the regular 70/10/20 partition for training, validation and testing.
We obtained a corpus of approximately 300,000 sentences, from which roughly 1.5 million single-quiz question training examples were derived.
|
What is the size of the dataset?
|
300,000 sentences with 1.5 million single-quiz questions
|
null | false
| null |
Write a friendly email to my teacher asking her to postpone our quiz schedule for tomorrow to next week because our basketball team is in the championship game tonight and all students will be out late celebrating our schools victory with pride!
|
Dear Mrs. Johnson,
I'm writing to request that you move tomorrows chemistry quiz to next week's class. As you know, our school basketball team is in the conference championship tonight. Most of the student body will either be in attendance or watching on TV (until late into the night.) We expect to win and to be celebrating our wonderful schools accomplishments. Given that this is such a great opportunity for our school to celebrate together, I think more students would attend and also do better on the quiz if it was postponed.
Respectfully,
Your best student Joe
|
|
null | false
| 217
|
Extractive multi-document summarization (MDS) aims to summarize a collection of documents by selecting a small number of sentences that represent the original content appropriately. Typical objectives for assembling a summary include information coverage and non-redundancy. A wide variety of methods have been introduced to approach MDS.
Many approaches are based on sentence ranking, i.e. assigning each sentence a score that indicates how well the sentence summarizes the input BIBREF0 , BIBREF1 , BIBREF2 . A summary is created by selecting the top entries of the ranked list of sentences. Since the sentences are often treated separately, these models might allow redundancy in the summary. Therefore, they are often extended by an anti-redundancy filter while de-queuing ranked sentence lists.
Other approaches work at summary-level rather than sentence-level and aim to optimize functions of sets of sentences to find good summaries, such as KL-divergence between probability distributions BIBREF3 or submodular functions that represent coverage, diversity, etc. BIBREF4
The centroid-based model belongs to the former group: it represents sentences as bag-of-word (BOW) vectors with TF-IDF weighting and uses a centroid of these vectors to represent the whole document collection BIBREF5 . The sentences are ranked by their cosine similarity to the centroid vector. This method is often found as a baseline in evaluations where it usually is outperformed BIBREF0 , BIBREF6 .
This baseline can easily be adapted to work at the summary-level instead the sentence level. This is done by representing a summary as the centroid of its sentence vectors and maximizing the similarity between the summary centroid and the centroid of the document collection. A simple greedy algorithm is used to find the best summary under a length constraint.
In order to keep the method efficient, we outline different methods to select a small number of candidate sentences from each document in the input collection before constructing the summary.
We test these modifications on the DUC2004 dataset for multi-document summarization. The results show an improvement of Rouge scores over the original centroid method. The performance is on par with state-of-the-art methods which shows that the similarity between a summary centroid and the input centroid is a well-suited function for global summary optimization.
The summarization approach presented in this paper is fast, unsupervised and simple to implement. Nevertheless, it performs as well as more complex state-of-the-art approaches in terms of Rouge scores on the DUC2004 dataset. It can be used as a strong baseline for future research or as a fast and easy-to-deploy summarization tool.
The results show an improvement of Rouge scores over the original centroid method.
|
Do the improvement of Rouge scores over the original centroid method?
|
Yes, they do.
|
null | false
| null |
The show follows Bluey, an anthropomorphic six-year-old Blue Heeler puppy who is characterised by her abundance of energy, imagination and curiosity of the world. The young dog lives with her father, Bandit; mother, Chilli; and younger sister, Bingo, who regularly joins Bluey on adventures as the pair embark on imaginative play together. Other characters featured each represent a different dog breed. Overarching themes include the focus on family, growing up and Australian culture. The program was created and is produced in Queensland; the culture and lifestyle of Brisbane inspires the show's setting.
|
Extract from the text the Australian city on which the show Bluey is inspired
|
Brisbane
|
null | false
| null |
On 7 December 1981 Grillo lost control of a Chevrolet K5 Blazer as he drove on a military road, forbidden to civilians, from Limone Piemonte to Colle di Tenda.
|
Without quoting directly from the text tell me how Beppe Grillo died.
|
He died in a car accident
|
null | false
| null |
What is the name of the gruff and tough female Viper pilot on the 2003 reboot of "Battlestar Galactica"?
|
Kara "Starbuck" Thrace
|
|
null | false
| null |
The announcement of the regulations which brought down the Wall took place at an hour-long press conference led by Günter Schabowski, the party leader in East Berlin and the top government spokesman, beginning at 18:00 CET on 9 November and broadcast live on East German television and radio. Schabowski was joined by Minister of Foreign Trade Gerhard Beil and Central Committee members Helga Labs and Manfred Banaschak.: 352
Schabowski had not been involved in the discussions about the new regulations and had not been fully updated. Shortly before the press conference, he was handed a note from Krenz announcing the changes, but given no further instructions on how to handle the information. The text stipulated that East German citizens could apply for permission to travel abroad without having to meet the previous requirements for those trips, and also allowed for permanent emigration between all border crossings—including those between East and West Berlin.
At 18:53, near the end of the press conference, ANSA's Riccardo Ehrman asked if the draft travel law of 6 November was a mistake. Schabowski gave a confusing answer that asserted it was necessary because West Germany had exhausted its capacity to accept fleeing East Germans, then remembered the note he had been given and added that a new regulation had been drafted to allow permanent emigration at any border crossing. This caused a stir in the room; amid several questions at once, Schabowski expressed surprise that the reporters had not yet seen this regulation, and started reading from the note. After this, a reporter, either Ehrman or Bild-Zeitung reporter Peter Brinkmann, both of whom were sitting in the front row at the press conference, asked when the regulations would take effect. After a few seconds' hesitation, Schabowski replied, "As far as I know, it takes effect immediately, without delay" (German: Das tritt nach meiner Kenntnis ... ist das sofort ... unverzüglich).: 352 This was an apparent assumption based on the note's opening paragraph; as Beil attempted to interject that it was up to the Council of Ministers to decide when it took effect, Schabowski proceeded to read this clause, which stated it was in effect until a law on the matter was passed by the Volkskammer. Crucially, a journalist then asked if the regulation also applied to the crossings to West Berlin. Schabowski shrugged and read item 3 of the note, which confirmed that it did.
After this exchange, Daniel Johnson of The Daily Telegraph asked what this law meant for the Berlin Wall. Schabowski sat frozen before giving a rambling statement about the Wall being tied to the larger disarmament question. He then ended the press conference promptly at 19:00 as journalists hurried from the room.
After the press conference, Schabowski sat for an interview with NBC News anchor Tom Brokaw in which he repeated that East Germans would be able to emigrate through the border and the regulations would go into effect immediately.
|
Please give me a short summary of the misinformed public announcements leading to the fall of the Berlin Wall
|
The announcement of the new laws which brought down the Berlin Wall took place at 6 pm on 9 November 1989, at a live press conference led by Günter Schabowski, the party leader in East Berlin. Schabowski had not been involved in the discussions about the new regulations and consequently was not adequately informed. Shortly before the press conference, he was handed a note outlining the changes to be announced, but given no further instructions on how to handle the content. The note specified that East German citizens could apply for permission to travel abroad, and permanent emigration between all border crossings—including those between East and West Berlin – would be allowed.
At 6:53 pm, near the end of the press conference, when asked if the draft travel law of 6 November was a mistake, Schabowski gave a confusing answer that it was necessary because West Germany had exhausted its capacity to accept fleeing East Germans, then remembered the note he had been handed and went on to add that a new regulation had been drafted to allow permanent emigration at any border crossing. This caused a stir in the room, several questions came flying at once. When asked when the regulations would take effect, Schabowski replied hesitantly, "As far as I know, it takes effect immediately, without delay". And after some confused interjections between party members, when asked what this law meant for the Berlin Wall, Schabowski froze for a few moments, before blurting out something to the effect of the Wall being tied to the larger disarmament question. He then ended the press conference promptly at 7 pm, as journalists hurried out of the room.
|
null | false
| null |
What are four creative ways to use a chess set besides playing chess?
|
You could use the pieces to invent a different war game; you could create a bottle flipping game with the king and queen; if the pieces are wood, you could use them as kindling for a Bo fire; and you could use the pieces as decorative paper weights.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.