Discovery and Separation of Features for Invariant Representation Learning

TL;DR

This paper introduces a novel framework for training neural networks that automatically discovers and separates predictive and nuisance factors, leading to invariant representations and improved generalization without needing nuisance labels.

Contribution

It presents an information theoretic approach to induce invariance by discovering and separating features, advancing beyond prior methods that require nuisance annotations.

Findings

Achieves state-of-the-art performance on various datasets.

Does not require nuisance annotations during training.

Enhances model robustness and generalization.

Abstract

Supervised machine learning models often associate irrelevant nuisance factors with the prediction target, which hurts generalization. We propose a framework for training robust neural networks that induces invariance to nuisances through learning to discover and separate predictive and nuisance factors of data. We present an information theoretic formulation of our approach, from which we derive training objectives and its connections with previous methods. Empirical results on a wide array of datasets show that the proposed framework achieves state-of-the-art performance, without requiring nuisance annotations during training.