Sparse and Structured Visual Attention
This addresses the issue of inefficient and less interpretable attention in multimodal tasks like VQA, offering incremental improvements for researchers and practitioners in computer vision and NLP.
The paper tackled the problem of softmax-based visual attention mechanisms assigning probability mass to all image regions, regardless of adjacency structure and relevance to text, by replacing them with sparsity-promoting transformations like sparsemax and a new TVmax method, resulting in gains in accuracy and higher similarity to human attention in VQA experiments.
Visual attention mechanisms are widely used in multimodal tasks, as visual question answering (VQA). One drawback of softmax-based attention mechanisms is that they assign some probability mass to all image regions, regardless of their adjacency structure and of their relevance to the text. In this paper, to better link the image structure with the text, we replace the traditional softmax attention mechanism with two alternative sparsity-promoting transformations: sparsemax, which is able to select only the relevant regions (assigning zero weight to the rest), and a newly proposed Total-Variation Sparse Attention (TVmax), which further encourages the joint selection of adjacent spatial locations. Experiments in VQA show gains in accuracy as well as higher similarity to human attention, which suggests better interpretability.