Don't Take the Easy Way Out: Ensemble Based Methods for Avoiding Known Dataset Biases

TL;DR

This paper introduces an ensemble-based training method that leverages biased models to improve robustness of AI models against dataset biases and domain shifts, demonstrated across multiple datasets.

Contribution

The paper proposes a two-stage ensemble training approach that uses biased models to enhance generalization in AI tasks, addressing dataset biases and out-of-domain challenges.

Findings

Significant robustness improvements on five datasets with out-of-domain tests.

12-point accuracy gain on a changing priors visual question answering dataset.

9-point accuracy gain on an adversarial question answering dataset.

Abstract

State-of-the-art models often make use of superficial patterns in the data that do not generalize well to out-of-domain or adversarial settings. For example, textual entailment models often learn that particular key words imply entailment, irrespective of context, and visual question answering models learn to predict prototypical answers, without considering evidence in the image. In this paper, we show that if we have prior knowledge of such biases, we can train a model to be more robust to domain shift. Our method has two stages: we (1) train a naive model that makes predictions exclusively based on dataset biases, and (2) train a robust model as part of an ensemble with the naive one in order to encourage it to focus on other patterns in the data that are more likely to generalize. Experiments on five datasets with out-of-domain test sets show significantly improved robustness in all settings, including a 12 point gain on a changing priors visual question answering dataset and a 9 point gain on an adversarial question answering test set.