Increasing Robustness to Spurious Correlations using Forgettable Examples

TL;DR

This paper introduces a method to improve NLP model robustness against spurious correlations by identifying minority examples through example forgetting and fine-tuning on these to enhance out-of-distribution generalization.

Contribution

The paper proposes using example forgetting to identify minority examples and a two-stage fine-tuning approach to boost model robustness against spurious correlations.

Findings

Significant improvements in out-of-distribution performance on MNLI, QQP, and FEVER datasets.

Effective identification of minority examples via example forgetting.

Enhanced model robustness through targeted fine-tuning on minority examples.

Abstract

Neural NLP models tend to rely on spurious correlations between labels and input features to perform their tasks. Minority examples, i.e., examples that contradict the spurious correlations present in the majority of data points, have been shown to increase the out-of-distribution generalization of pre-trained language models. In this paper, we first propose using example forgetting to find minority examples without prior knowledge of the spurious correlations present in the dataset. Forgettable examples are instances either learned and then forgotten during training or never learned. We empirically show how these examples are related to minorities in our training sets. Then, we introduce a new approach to robustify models by fine-tuning our models twice, first on the full training data and second on the minorities only. We obtain substantial improvements in out-of-distribution generalization when applying our approach to the MNLI, QQP, and FEVER datasets.