BHN: A Brain-like Heterogeneous Network

TL;DR

This paper introduces BHN, a brain-inspired heterogeneous network that learns distributed and global representations through self-supervised, gradient-isolated optimization, enhancing image and video understanding.

Contribution

It presents a novel brain-like architecture that cooperatively learns multiple representations using a minimax optimization strategy, inspired by brain functions.

Findings

Improved representation quality in image and video tasks

Effective self-supervised learning with gradient isolation

Demonstrated advantages over traditional models

Abstract

The human brain works in an unsupervised way, and more than one brain region is essential for lighting up intelligence. Inspired by this, we propose a brain-like heterogeneous network (BHN), which can cooperatively learn a lot of distributed representations and one global attention representation. By optimizing distributed, self-supervised, and gradient-isolated objective functions in a minimax fashion, our model improves its representations, which are generated from patches of pictures or frames of videos in experiments.