Videos

Consumer health questions pose specific challenges to automated answering. Two of the salient aspects are the higher linguistic and semantic complexity when compared to open domain questions, and the more pronounced need for reliable information. In this talk I will present two main approaches to deal with the increased complexity by recognizing question entailment and by question summarization, recently published respectively in BMC Bioinformatics and ACL 2019. In particular, our question entailment approach to question answering (QA) showed that restricting the answer sources to only reliable resources led to an improvement of the QA performance and our summarization experiments showed the relevance of data augmentation methods for abstractive question summarization. I’ll also talk about the MEDIQA shared task on question entailment, textual inference and medical question answering that we recently organized at ACL-BioNLP. In the second part of the talk, I will address more specifically questions about medications and present our last study and dataset on medication QA. Finally, I’ll describe our recent endeavors in visual question answering (VQA) from radiology images and the medical VQA challenge (VQA-Med) editions for 2019 and 2020 that we organize in the scope of ImageCLEF.

The explosion of available idea repositories -- scientific papers, patents, product descriptions -- represents an unprecedented opportunity to accelerate innovation and lead to a wealth of discoveries. Given the scale of the problem and its ever-expanding nature, there is a need for intelligent automation to assist in the process of discovery. In this talk, I will present our work toward addressing this challenging problem. We developed an approach for boosting people’s creativity by helping them discover analogies -- abstract structural connections between ideas. We learn to decompose innovation texts into functional models that describe the components and goals of inventions, and use them to build a search engine supporting expressive inspiration queries. In ideation studies, our inspirations helped people generate better ideas with significant improvement over standard search. We also construct a commonsense ontology of purposes and mechanisms of products, mapping the landscape of ideas. I will also describe a novel machine learning framework we developed in order to identify innovation in patents, where labels are extremely hard to obtain. In our setting, called Ballpark Learning, we are only given groups of instances with coarse constraints over label averages. We demonstrate encouraging results in classification and regression tasks across several domains.

Many promising cancer immunotherapy treatment protocols rely on efficient and increasingly extensive methods for manipulating human immune cells. T cells are a frequent target of the laboratory and clinical research driving the development of such protocols as they are most often the effector of the cytotoxic activity that makes these treatments so potent. However, the cytokine signaling network that drives the differentiation and function of such cells is complex and difficult to replicate on a large scale in model biological systems. Abridged versions of these networks have been established over decades of research but it remains challenging to define their global structure as the classification of T cell subtypes operating in these networks, the mechanics of their formation, and the purpose of the signaling molecules they excrete are all controversial, with a slowly expanding understanding emerging in literature over time. To aid in the quantification of this understanding, we are developing a methodology for identifying references to well known cytokines, transcription factors, and T cell types in literature as well as classifying the relationships between the three in an attempt to determine what cytokines initiate the transcription programs that lead to various cell states in addition to the secretion profiles associated with those states. Entity recognition for this task is performed using SciSpacy and classification of the relations between these entities is based on an LSTM trained using Snorkel, where weak supervision is established through a variety of classification heuristics and distant supervision is provided via previously published immunology databases.

Consider the difference between the two sentences “Pat didn’t remember to water the plants” and “Pat didn’t remember that she had watered the plants.” Fluent English speakers recognize that the former sentence implies that Pat did not water the plants, while the latter sentence implies she did. This distinction is crucial to understanding the meaning of these sentences, yet it is one that automated natural language processing (NLP) systems struggle to make. In this talk, I will discuss my work on developing state-of-the-art NLP models that make essential inferences about events (e.g., a “watering” event) and participants (e.g., “Pat” and “the plants”) in natural language sentences. In particular, I will focus on two supervised NLP tasks that serve as core tests of language understanding: Event Factuality Prediction and Semantic Proto-Role Labeling. I will also discuss my work on unsupervised acquisition of common-sense knowledge from large natural language text corpora, and the concomitant challenge of detecting problematic social biases in NLP models trained on such data.

We present two results: 1) Analysis techniques for state-of-the-art question-answering models on images, tables and passages of text. We show how these networks often ignore important question terms. Leveraging such non-robust behavior, we present a variety of adversarial examples derived by perturbing the questions. Our strongest attacks drop the accuracy of a visual question answering model from 61.1% to 19%, and that of a tabular question answering model from 33.5% to 3.3%. We demonstrate that attributions can augment standard measures of accuracy and empower investigation of model performance. When a model is accurate but for the wrong reasons, attributions can surface erroneous logic in the model that indicates inadequacies in the data. 2) Parameter-efficient transfer learning: We present a novel method for re-purposing pretrained neural networks to new tasks while maintaining most of the weights intact. The basic approach is to learn a model patch - a small set of parameters - that will specialize to each task, instead of fine-tuning the last layer or the entire network. For instance, we show that learning a set of scales and biases is sufficient to convert a pretrained network to perform well on qualitatively different problems (e.g. converting a Single Shot MultiBox Detection (SSD) model into a 1000-class image classification model while reusing 98% of parameters of the SSD feature extractor). Our approach allows both simultaneous (multi-task) as well as sequential transfer learning. In several multi-task learning problems, despite using much fewer parameters than traditional logits-only fine-tuning, we match single-task performance.

A longstanding goal of artificial intelligence (AI) is to develop agents that can assist or augment humans. Such agents have the potential to transform society. While AI agents can excel at well-defined tasks like games, much more limited progress has been made solving real-world problems like interacting with humans, where data collection is costly, objectives are ill-defined, and safety is critical. In this talk, I will discuss how we can design agents to improve the efficiency and success of collective human work ("crowdsourcing"), by leveraging techniques from AI, reinforcement learning, and optimization, together with structured contributions from human workers and task designers. This approach improves on current methods for designing such agents, which typically require large amounts of manual experimentation and costly data collection to get right. I will demonstrate the effectiveness of this approach on several crowdsourcing management problems, and also share recent work on how agents can make shared decisions with humans to achieve better outcomes.

From robots to cars, virtual assistants and voice-controlled drones, computing devices are increasingly expected to communicate naturally with people and to understand the visual context in which they operate. In this talk, I will present our latest work on generating and comprehending visually-grounded language. First, we will discuss the challenging task of describing an image (image captioning). I will introduce captioning models that leverage multiple data sources, including object detection datasets and unaligned text corpora, in order to learn about the long-tail of visual concepts found in the real world. To support and encourage further efforts in this area, I will present the 'nocaps' benchmark for novel object captioning. In the second part of the talk, I will describe our recent work on developing agents that follow natural language instructions in reconstructed 3D environments using the R2R dataset for vision-and-language navigation.

The daily actions and decisions of people are increasingly shaped by recommendation systems, from e-commerce and content platforms to education and wellness applications. These systems selectively suggest and present information items based on their characterization of user preferences. However, existing preference modeling methods are limited due to the incomplete and biased nature of the behavioral data that inform the models. As a result, recommendations can be narrow, skewed, homogeneous, and divergent from users’ aspirations. In this talk, I will introduce user-centric recommendation models and systems that address the incompleteness and bias of existing methods and increase systems’ utility for individuals. Specifically, I will present my work addressing two key research challenges: (1) inferring debiased preferences from biased behavioral data using counterfactual reasoning, and (2) eliciting unobservable current and aspirational preferences from users through interactive machine learning. I will conclude with discussion of field experiments that demonstrate how user-centric systems can promote healthier diets and better content choices.

As the availability of data for language learning grows, the role of linguistic structure is under scrutiny. At the same time, it is imperative to closely inspect patterns in data which might present loopholes for models to obtain high performance on benchmarks. In a two-part talk, I will address each of these challenges. First, I will introduce the paradigm of scaffolded learning. Scaffolds enable us to leverage inductive biases from one structural source for prediction of a different, but related structure, using only as much supervision as is necessary. We show that the resulting representations achieve improved performance across a range of tasks, indicating that linguistic structure remains beneficial even with powerful deep learning architectures. In the second part of the talk, I will showcase some of the properties exhibited by NLP models in large data regimes. Even as these models report excellent performance, sometimes claimed to beat humans, a closer look reveals that predictions are not a result of complex reasoning, and the task is not being completed in a generalizable way. Instead, this success can be largely attributed to exploitation of some artifacts of annotation in the datasets. I will discuss some questions our finding raises, as well as directions for future work.