Learning from others' mistakes: Avoiding dataset biases without modeling them

TL;DR

This paper introduces a method for training NLP models that automatically avoid dataset biases by leveraging errors from limited-capacity models, improving robustness without explicitly modeling biases.

Contribution

The paper proposes a bias-agnostic training approach using a product of experts, which does not require prior knowledge of specific dataset biases.

Findings

Improves out-of-distribution performance

Effective without explicit bias identification

Leverages errors of limited-capacity models

Abstract

State-of-the-art natural language processing (NLP) models often learn to model dataset biases and surface form correlations instead of features that target the intended underlying task. Previous work has demonstrated effective methods to circumvent these issues when knowledge of the bias is available. We consider cases where the bias issues may not be explicitly identified, and show a method for training models that learn to ignore these problematic correlations. Our approach relies on the observation that models with limited capacity primarily learn to exploit biases in the dataset. We can leverage the errors of such limited capacity models to train a more robust model in a product of experts, thus bypassing the need to hand-craft a biased model. We show the effectiveness of this method to retain improvements in out-of-distribution settings even if no particular bias is targeted by the biased model.