Unsupervised Adversarial Invariance

TL;DR

This paper introduces an unsupervised adversarial framework that learns invariant data representations without labeled nuisance factors, improving generalization and domain adaptation in neural networks.

Contribution

It proposes a novel unsupervised method for invariance induction using adversarial training, eliminating the need for labeled nuisance factors or domain knowledge.

Findings

Outperforms supervised methods in inducing invariance

Effective in domain adaptation tasks

Utilizes synthetic data augmentation for invariance learning

Abstract

Data representations that contain all the information about target variables but are invariant to nuisance factors benefit supervised learning algorithms by preventing them from learning associations between these factors and the targets, thus reducing overfitting. We present a novel unsupervised invariance induction framework for neural networks that learns a split representation of data through competitive training between the prediction task and a reconstruction task coupled with disentanglement, without needing any labeled information about nuisance factors or domain knowledge. We describe an adversarial instantiation of this framework and provide analysis of its working. Our unsupervised model outperforms state-of-the-art methods, which are supervised, at inducing invariance to inherent nuisance factors, effectively using synthetic data augmentation to learn invariance, and domain adaptation. Our method can be applied to any prediction task, eg., binary/multi-class classification or regression, without loss of generality.