Bidirectional Beam Search: Forward-Backward Inference in Neural Sequence Models for Fill-in-the-Blank Image Captioning
This addresses a bottleneck in using bidirectional models for sequence decoding, enabling improved performance in tasks like image captioning that require reasoning about past and future context.
The paper tackles the problem of approximate inference in bidirectional neural sequence models by introducing Bidirectional Beam Search (BiBS), an algorithm that extends beam search to handle both forward and backward dependencies, and demonstrates its effectiveness on a novel Fill-in-the-Blank Image Captioning task, consistently outperforming baseline methods.
We develop the first approximate inference algorithm for 1-Best (and M-Best) decoding in bidirectional neural sequence models by extending Beam Search (BS) to reason about both forward and backward time dependencies. Beam Search (BS) is a widely used approximate inference algorithm for decoding sequences from unidirectional neural sequence models. Interestingly, approximate inference in bidirectional models remains an open problem, despite their significant advantage in modeling information from both the past and future. To enable the use of bidirectional models, we present Bidirectional Beam Search (BiBS), an efficient algorithm for approximate bidirectional inference.To evaluate our method and as an interesting problem in its own right, we introduce a novel Fill-in-the-Blank Image Captioning task which requires reasoning about both past and future sentence structure to reconstruct sensible image descriptions. We use this task as well as the Visual Madlibs dataset to demonstrate the effectiveness of our approach, consistently outperforming all baseline methods.