Reasoning-Modulated Representations

TL;DR

This paper explores how incorporating reasoning modules with prior knowledge into neural networks can improve the quality of learned representations, especially in physics-informed tasks, advancing self-supervised learning methods.

Contribution

It introduces a novel approach that integrates reasoning modules with prior knowledge into neural networks to shape representations in self-supervised learning.

Findings

Reasoning modules influence learned representations positively.

Incorporating priors reduces data requirements for training.

Approach enhances generalization in physics-based tasks.

Abstract

Neural networks leverage robust internal representations in order to generalise. Learning them is difficult, and often requires a large training set that covers the data distribution densely. We study a common setting where our task is not purely opaque. Indeed, very often we may have access to information about the underlying system (e.g. that observations must obey certain laws of physics) that any "tabula rasa" neural network would need to re-learn from scratch, penalising performance. We incorporate this information into a pre-trained reasoning module, and investigate its role in shaping the discovered representations in diverse self-supervised learning settings from pixels. Our approach paves the way for a new class of representation learning, grounded in algorithmic priors.