Paper reading: Learning Visual Question Answering by Bootstrapping hard Attention

Learning Visual Question Answering by Bootstrapping hard Attention

Google DeepMind ECCV-2018

2018-08-05 19:24:44

Paper:https://arxiv.org/abs/1808.00300

Introduction:

This article attempts to use hard attention method to key out the most useful feature, to do VQA task learning.

Soft Attention:　　

Existing attention models [7,8,9,10] be predominantly based on soft attention, in which all information is adaptively Re-weighted before being aggregated. This can improve accuracy by isolating important information and avoiding interference from unimportant information.

Hard Attention:

--

Paper reading: Learning Visual Question Answering by Bootstrapping hard Attention