Near-Optimal Glimpse Sequences for Training Hard Attention Neural NetworksDownload PDF

Sep 28, 2020 (edited Mar 05, 2021)ICLR 2021 Conference Blind SubmissionReaders: Everyone
  • Reviewed Version (pdf):
  • Keywords: attention, hard attention, variational inference, bayesian optimal experimental design
  • Abstract: Hard visual attention is a promising approach to reduce the computational burden of modern computer vision methodologies. Hard attention mechanisms are typically non-differentiable. They can be trained with reinforcement learning but the high-variance training this entails hinders more widespread application. We show how hard attention for image classification can be framed as a Bayesian optimal experimental design (BOED) problem. From this perspective, the optimal locations to attend to are those which provide the greatest expected reduction in the entropy of the classification distribution. We introduce methodology from the BOED literature to approximate this optimal behaviour, and use it to generate `near-optimal' sequences of attention locations. We then show how to use such sequences to partially supervise, and therefore speed up, the training of a hard attention mechanism. Although generating these sequences is computationally expensive, they can be reused by any other networks later trained on the same task.
  • One-sentence Summary: We use Bayesian experimental design to produce sequences which are later used to provide a supervision signal for a hard attention network and greatly speed up its training.
  • Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics
  • Supplementary Material: zip
15 Replies