Latent Alignment and Variational Attention
This addresses a problem for researchers and practitioners in NLP and related fields by providing a more probabilistic and trainable attention method, though it is incremental as it builds on existing attention and variational inference techniques.
The paper tackled the issue that neural attention does not marginalize over latent alignments probabilistically, making it hard to compare or compose with other models, and proposed variational attention networks as an alternative. Experiments showed that variational attention retains most performance gains of exact latent models for tasks like machine translation and visual question answering, with training speed comparable to neural attention.
Neural attention has become central to many state-of-the-art models in natural language processing and related domains. Attention networks are an easy-to-train and effective method for softly simulating alignment; however, the approach does not marginalize over latent alignments in a probabilistic sense. This property makes it difficult to compare attention to other alignment approaches, to compose it with probabilistic models, and to perform posterior inference conditioned on observed data. A related latent approach, hard attention, fixes these issues, but is generally harder to train and less accurate. This work considers variational attention networks, alternatives to soft and hard attention for learning latent variable alignment models, with tighter approximation bounds based on amortized variational inference. We further propose methods for reducing the variance of gradients to make these approaches computationally feasible. Experiments show that for machine translation and visual question answering, inefficient exact latent variable models outperform standard neural attention, but these gains go away when using hard attention based training. On the other hand, variational attention retains most of the performance gain but with training speed comparable to neural attention.