Are Disentangled Representations Helpful for Abstract Visual Reasoning?

TL;DR

This study empirically investigates whether disentangled representations improve performance on abstract reasoning tasks, finding that they enable faster learning and better results across multiple models.

Contribution

The paper provides large-scale empirical evidence that disentangled representations enhance abstract reasoning performance and sample efficiency.

Findings

Disentangled representations lead to better reasoning performance.

They enable quicker learning with fewer samples.

Empirical validation across 360 models and tasks.

Abstract

A disentangled representation encodes information about the salient factors of variation in the data independently. Although it is often argued that this representational format is useful in learning to solve many real-world down-stream tasks, there is little empirical evidence that supports this claim. In this paper, we conduct a large-scale study that investigates whether disentangled representations are more suitable for abstract reasoning tasks. Using two new tasks similar to Raven's Progressive Matrices, we evaluate the usefulness of the representations learned by 360 state-of-the-art unsupervised disentanglement models. Based on these representations, we train 3600 abstract reasoning models and observe that disentangled representations do in fact lead to better down-stream performance. In particular, they enable quicker learning using fewer samples.