Adversarial Policy Gradient for Learning Graph-Based Representation in Human Visual Processing

Subhrasankar Chatterjee, Subrata Pain, Debasis Samanta

Keywords: Graph Representation, Voxelwise Visual Modelling, Reinforcement Learning, Adversarial Game

TL;DR: This article proposes a novel approach for constructing a graph-based representation of brain fMRI data to facilitate Visual Encoding and Decoding

Abstract:

This article discusses the challenges in modeling the neural mechanisms underlying human visual processing and the use of graph-based representations to capture inter-region relationships in visual processing. While graphs have shown promise in analyzing neural responses, learning an optimal graph representation from limited data is challenging, as there is no ground truth to learn from. To address this,
the authors propose a novel approach to graph-based representation using Adversarial Policy Gradient, which involves an adversarial game between the Policy Network and the Reward Network to generate a better graph representation.