Explaining Representation Learning with Perceptual Components

TL;DR

This paper introduces a novel interpretability method for self-supervised models by analyzing their representation spaces through perceptual components like color, shape, and texture, enhancing human-aligned explanations.

Contribution

It proposes a new approach using perceptual component masking to interpret representation spaces, addressing the challenge of explainability in self-supervised learning.

Findings

Importance maps reveal perceptual component significance in representations.

Different training objectives produce distinct perceptual component representations.

The method provides intuitive explanations aligned with human visual perception.

Abstract

Self-supervised models create representation spaces that lack clear semantic meaning. This interpretability problem of representations makes traditional explainability methods ineffective in this context. In this paper, we introduce a novel method to analyze representation spaces using three key perceptual components: color, shape, and texture. We employ selective masking of these components to observe changes in representations, resulting in distinct importance maps for each. In scenarios, where labels are absent, these importance maps provide more intuitive explanations as they are integral to the human visual system. Our approach enhances the interpretability of the representation space, offering explanations that resonate with human visual perception. We analyze how different training objectives create distinct representation spaces using perceptual components. Additionally, we examine the representation of images across diverse image domains, providing insights into the role of these components in different contexts.