Agree to Disagree: Diversity through Disagreement for Better Transferability
This addresses the OOD generalization problem in machine learning, which is crucial for real-world applications where training and test distributions differ, but it is incremental as it builds on ensemble and disagreement-based methods.
The paper tackles the problem of limited model diversity in gradient-based learning, which hinders transferability by favoring spurious features and using only a subset of predictive features, especially in out-of-distribution (OOD) scenarios. It proposes D-BAT, an algorithm that enforces agreement on training data and disagreement on OOD data, showing improvements in mitigating shortcut-learning, enhancing uncertainty and OOD detection, and boosting transferability in experiments.
Gradient-based learning algorithms have an implicit simplicity bias which in effect can limit the diversity of predictors being sampled by the learning procedure. This behavior can hinder the transferability of trained models by (i) favoring the learning of simpler but spurious features -- present in the training data but absent from the test data -- and (ii) by only leveraging a small subset of predictive features. Such an effect is especially magnified when the test distribution does not exactly match the train distribution -- referred to as the Out of Distribution (OOD) generalization problem. However, given only the training data, it is not always possible to apriori assess if a given feature is spurious or transferable. Instead, we advocate for learning an ensemble of models which capture a diverse set of predictive features. Towards this, we propose a new algorithm D-BAT (Diversity-By-disAgreement Training), which enforces agreement among the models on the training data, but disagreement on the OOD data. We show how D-BAT naturally emerges from the notion of generalized discrepancy, as well as demonstrate in multiple experiments how the proposed method can mitigate shortcut-learning, enhance uncertainty and OOD detection, as well as improve transferability.