How saccadic vision might help with theinterpretability of deep networks

TL;DR

This paper proposes a biologically inspired saccadic vision model to improve interpretability and object recognition in deep neural networks, supported by initial experimental results.

Contribution

It introduces a novel saccadic perception mechanism for deep networks, addressing interpretability and object-orientedness issues.

Findings

Proof of concept experimental results support the approach

Saccadic mechanism enhances interpretability of deep networks

Potential for improved object recognition in neural models

Abstract

We describe how some problems (interpretability,lack of object-orientedness) of modern deep networks potentiallycould be solved by adapting a biologically plausible saccadicmechanism of perception. A sketch of such a saccadic visionmodel is proposed. Proof of concept experimental results areprovided to support the proposed approach.