Multi-Mention Learning for Reading Comprehension with Neural Cascades

Reading comprehension is a challenging task, especially when executed across longer or across multiple evidence documents, where the answer is likely to reoccur. Existing neural architectures typically do not scale to the entire evidence, and hence, resort to selecting a single passage in the document (either via truncation or other means), and carefully searching for the answer within that passage. However, in some cases, this strategy can be suboptimal, since by focusing on a specific passage, it becomes difficult to leverage multiple mentions of the same answer throughout the document. In this work, we take a different approach by constructing lightweight models that are combined in a cascade to find the answer. Each submodel consists only of feed-forward networks equipped with an attention mechanism, making it trivially parallelizable. We show that our approach can scale to approximately an order of magnitude larger evidence documents and can aggregate information at the representation level from multiple mentions of each answer candidate across the document. Empirically, our approach achieves state-of-the-art performance on both the Wikipedia and web domains of the TriviaQA dataset, outperforming more complex, recurrent architectures.

A Decomposable Attention Model for Natural Language Inference

We propose a simple neural architecture for natural language inference. Our approach uses attention to decompose the problem into subproblems that can be solved separately, thus making it trivially parallelizable. On the Stanford Natural Language Inference (SNLI) dataset, we obtain state-of-the-art results with almost an order of magnitude fewer parameters than previous work and without relying on any word-order information. Adding intra-sentence attention that takes a minimum amount of order into account yields further improvements.

Language Modeling with Power Low Rank Ensembles

We present power low rank ensembles (PLRE), a flexible framework for n-gram language modeling where ensembles of low rank matrices and tensors are used to obtain smoothed probability estimates of words in context. Our method can be understood as a generalization of n-gram modeling to non-integer n, and includes standard techniques such as absolute discounting and Kneser-Ney smoothing as special cases. PLRE training is efficient and our approach outperforms state-of-the-art modified Kneser Ney baselines in terms of perplexity on large corpora as well as on BLEU score in a downstream machine translation task.

Infinite Mixed Membership Matrix Factorization

Rating and recommendation systems have become a popular application area for applying a suite of machine learning techniques. Current approaches rely primarily on probabilistic interpretations and extensions of matrix factorization, which factorizes a user-item ratings matrix into latent user and item vectors. Most of these methods fail to model significant variations in item ratings from otherwise similar users, a phenomenon known as the "Napoleon Dynamite" effect. Recent efforts have addressed this problem by adding a contextual bias term to the rating, which captures the mood under which a user rates an item or the context in which an item is rated by a user. In this work, we extend this model in a nonparametric sense by learning the optimal number of moods or contexts from the data, and derive Gibbs sampling inference procedures for our model. We evaluate our approach on the MovieLens 1M dataset, and show significant improvements over the optimal parametric baseline, more than twice the improvements previously encountered for this task. We also extract and evaluate a DBLP dataset, wherein we predict the number of papers co-authored by two authors, and present improvements over the parametric baseline on this alternative domain as well.

A Spectral Algorithm for Latent Junction Trees

Latent variable models are an elegant framework for capturing rich probabilistic dependencies in many applications. However, current approaches typically parametrize these models using conditional probability tables, and learning relies predominantly on local search heuristics such as Expectation Maximization. Using tensor algebra, we propose an alternative parameterization of latent variable models (where the model structures are junction trees) that still allows for computation of marginals among observed variables. While this novel representation leads to a moderate increase in the number of parameters for junction trees of low treewidth, it lets us design a local-minimum-free algorithm for learning this parameterization. The main computation of the algorithm involves only tensor operations and SVDs which can be orders of magnitude faster than EM algorithms for large datasets. To our knowledge, this is the first provably consistent parameter learning technique for a large class of low-treewidth latent graphical models beyond trees. We demonstrate the advantages of our method on synthetic and real datasets.

Infinite Hierarchical MMSB Model for Nested Communities/Groups in Social Networks

Actors in realistic social networks play not one but a number of diverse roles depending on whom they interact with, and a large number of such role-specific interactions collectively determine social communities and their organizations. Methods for analyzing social networks should capture these multi-faceted role-specific interactions, and, more interestingly, discover the latent organization or hierarchy of social communities. We propose a hierarchical Mixed Membership Stochastic Blockmodel to model the generation of hierarchies in social communities, selective membership of actors to subsets of these communities, and the resultant networks due to within- and cross-community interactions. Furthermore, to automatically discover these latent structures from social networks, we develop a Gibbs sampling algorithm for our model. We conduct extensive validation of our model using synthetic networks, and demonstrate the utility of our model in real-world datasets such as predator-prey networks and citation networks.