With the devastating outbreak of COVID-19, vaccines are one of the crucial lines of defense against mass infection in this global pandemic. Given the protection they provide, vaccines are becoming mandatory in certain social and professional settings. This paper presents a classification model for detecting COVID-19 vaccination related search queries, a machine learning model that is used to generate search insights for COVID-19 vaccinations. The proposed method combines and leverages advancements from modern state-of-the-art (SOTA) natural language understanding (NLU) techniques such as pretrained Transformers with traditional dense features. We propose a novel approach of considering dense features as memory tokens that the model can attend to. We show that this new modeling approach enables a significant improvement to the Vaccine Search Insights (VSI) task, improving a strong well-established gradient-boosting baseline by relative +15% improvement in F1 score and +14% in precision.

Language models are widely deployed to provide automatic text completion services in user products. However, recent research has revealed that language models (especially large ones) bear considerable risk of memorizing private training data, which is then vulnerable to leakage and extraction by adversaries. In this study, we test the efficacy of a range of privacy-preserving techniques to mitigate unintended memorization of sensitive user text, while varying other factors such as model size and adversarial conditions. We test both "heuristic" mitigations (those without formal privacy guarantees) and Differentially Private training, which provides provable levels of privacy at the cost of some model performance. Our experiments show that (with the exception of L2 regularization), heuristic mitigations are largely ineffective in preventing memorization in our test suite, possibly because they make too strong of assumptions about the characteristics that define "sensitive" or "private" text. In contrast, Differential Privacy reliably prevents memorization in our experiments, despite its computational and model-performance costs.

Classification algorithms using Transformer architectures can be affected by the sequence length learning problem whenever observations from different classes have a different length distribution. This problem brings models to use sequence length as a predictive feature instead of relying on important textual information. Even if most public datasets are not affected by this problem, privately corpora for fields such as medicine and insurance may carry this data bias. This poses challenges throughout the value chain given their usage in a machine learning application. In this paper, we empirically expose this problem and present approaches to minimize its impacts.

A widely acknowledged shortcoming of WordNet is that it lacks a distinction between word meanings which are systematically related (polysemy), and those which are coincidental (homonymy). Several previous works have attempted to fill this gap, by inferring this information using computational methods. We revisit this task, and exploit recent advances in language modelling to synthesise homonymy annotation for Princeton WordNet. Previous approaches treat the problem using clustering methods; by contrast, our method works by linking WordNet to the Oxford English Dictionary, which contains the information we need. To perform this alignment, we pair definitions based on their proximity in an embedding space produced by a Transformer model. Despite the simplicity of this approach, our best model attains an F1 of .97 on an evaluation set that we annotate. The outcome of our work is a high-quality homonymy annotation layer for Princeton WordNet, which we release.

We introduce hp-greedy, a refinement approach for building gravitational wave surrogates as an extension of the standard reduced basis framework. Our proposal is data-driven, with a domain decomposition of the parameter space, local reduced basis, and a binary tree as the resulting structure, which are obtained in an automated way. When compared to the standard global reduced basis approach, the numerical simulations of our proposal show three salient features: i) representations of lower dimension with no loss of accuracy, ii) a significantly higher accuracy for a fixed maximum dimensionality of the basis, in some cases by orders of magnitude, and iii) results that depend on the reduced basis seed choice used by the refinement algorithm. We first illustrate the key parts of our approach with a toy model and then present a more realistic use case of gravitational waves emitted by the collision of two spinning, non-precessing black holes. We discuss performance aspects of hp-greedy, such as overfitting with respect to the depth of the tree structure, and other hyperparameter dependences. As two direct applications of the proposed hp-greedy refinement, we envision: i) a further acceleration of statistical inference, which might be complementary to focused reduced-order quadratures, and ii) the search of gravitational waves through clustering and nearest neighbors.

Artificial intelligence and natural language processing (NLP) are increasingly being used in customer service to interact with users and answer their questions. The goal of this systematic review is to examine existing research on the use of NLP technology in customer service, including the research domain, applications, datasets used, and evaluation methods. The review also looks at the future direction of the field and any significant limitations. The review covers the time period from 2015 to 2022 and includes papers from five major scientific databases. Chatbots and question-answering systems were found to be used in 10 main fields, with the most common use in general, social networking, and e-commerce areas. Twitter was the second most commonly used dataset, with most research also using their own original datasets. Accuracy, precision, recall, and F1 were the most common evaluation methods. Future work aims to improve the performance and understanding of user behavior and emotions, and address limitations such as the volume, diversity, and quality of datasets. This review includes research on different spoken languages and models and techniques.

Pre-trained language models (LMs) have shown remarkable reasoning performance using explanations (or ``chain-of-thought'' (CoT)) for in-context learning. On the other hand, these reasoning tasks are usually presumed to be more approachable for symbolic programming. To make progress towards understanding in-context learning, we curate synthetic datasets containing equivalent (natural, symbolic) data pairs, where symbolic examples contain first-order logic rules and predicates from knowledge bases (KBs). Then we revisit neuro-symbolic approaches and use Language Models as Logic Programmer (LMLP) that learns from demonstrations containing logic rules and corresponding examples to iteratively reason over KBs, recovering Prolog's backward chaining algorithm. Comprehensive experiments are included to systematically compare LMLP with CoT in deductive reasoning settings, showing that LMLP enjoys more than 25% higher accuracy than CoT on length generalization benchmarks even with fewer parameters.

Estimating treatment effects is one of the most challenging and important tasks of data analysts. In many applications, like online marketing and personalized medicine, treatment needs to be allocated to the individuals where it yields a high positive treatment effect. Uplift models help select the right individuals for treatment and maximize the overall treatment effect (uplift). A major challenge in uplift modeling concerns model evaluation. Previous literature suggests methods like the Qini curve and the transformed outcome mean squared error. However, these metrics suffer from variance: their evaluations are strongly affected by random noise in the data, which renders their signals, to a certain degree, arbitrary. We theoretically analyze the variance of uplift evaluation metrics and derive possible methods of variance reduction, which are based on statistical adjustment of the outcome. We derive simple conditions under which the variance reduction methods improve the uplift evaluation metrics and empirically demonstrate their benefits on simulated and real-world data. Our paper provides strong evidence in favor of applying the suggested variance reduction procedures by default when evaluating uplift models on RCT data.

Headquartered in San Francisco, Tubi is an ad-supported video-on-demand (AVOD) service with movies and television shows. With over 40,000 titles from every major Hollywood studio, Tubi gives fans of movies and television shows an easy way to discover new content that is available completely free. Tubi's library has something for every member of our diverse audience, and we're committed to building a workforce that reflects that diversity. We're looking for great people who are creative thinkers, self-motivators, and impact-makers looking to help shape the future of streaming. Our services are currently available in the US, Canada, Australia, New Zealand, Mexico, Costa Rica, Ecuador, El Salvador, Guatemala, and Panama.

Lara Croft is making a return with the help of Amazon Games. The company will publish the next Tomb Raider title, which Crystal Dynamics is developing. Amazon said the studio is using Unreal Engine 5 to craft the biggest and most expansive Tomb Raider game to date. It'll likely be a few years before you can get your hands on the next Tomb Raider game, which is coming to multiple platforms. It's still in early development, but Amazon says it'll retain all of the franchise's hallmarks.