Visual Attention Through Uncertainty Minimization in Recurrent Generative Models
Keywords
No Thumbnail Available
Authors
Issue Date
2019-08-14
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Allocating visual attention through saccadic eye movements is a key ability of intelligent
agents. Attention is both influenced through bottom-up stimulus properties as well as topdown
task demands. The interaction of these two attention mechanisms is not yet fully
understood. A parsimonious reconciliation posits that both processes serve the minimization
of predictive uncertainty. We propose a recurrent generative neural network model that
predicts a visual scene based on foveated glimpses. The model shifts its attention in order
to minimize the uncertainty in its predictions. We show that the proposed model produces
naturalistic eye-movements focusing on salient stimulus regions. Introducing the additional
task of classifying the stimulus modulates the saccade patterns and enables effective image
classification. Given otherwise equal conditions, we show that different task requirements
cause the model to focus on distinct, task-relevant regions. The model’s saccade statistics
correspond well with previous experimental data in humans and provide insights into
unsettled controversies in the literature. The results provide evidence that uncertainty
minimization could be a fundamental mechanisms for the allocation of visual attention.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen