Discovery and Separation of Features for Invariant Representation Learning
This addresses generalization issues in machine learning by improving robustness to nuisances, but it is incremental as it builds on prior invariance methods.
The paper tackles the problem of supervised models associating irrelevant nuisance factors with prediction targets, which harms generalization, by proposing a framework that learns to discover and separate predictive and nuisance factors for invariant representation learning, achieving state-of-the-art performance across multiple datasets without needing nuisance annotations.
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.