Not Eliminate but Aggregate: Post-Hoc Control over Mixture-of-Experts to Address Shortcut Shifts in Natural Language Understanding

TL;DR

This paper proposes a post-hoc aggregation method for mixture-of-experts models to improve robustness against shortcut reliance and distribution shifts in natural language understanding tasks.

Contribution

It introduces a novel post-hoc control technique for mixture-of-experts to mitigate shortcut dependence without retraining the entire model.

Findings

Enhanced robustness to shortcut-induced distribution shifts

Post-hoc aggregation improves model reliability

Practical advantages over previous elimination methods

Abstract

Recent models for natural language understanding are inclined to exploit simple patterns in datasets, commonly known as shortcuts. These shortcuts hinge on spurious correlations between labels and latent features existing in the training data. At inference time, shortcut-dependent models are likely to generate erroneous predictions under distribution shifts, particularly when some latent features are no longer correlated with the labels. To avoid this, previous studies have trained models to eliminate the reliance on shortcuts. In this study, we explore a different direction: pessimistically aggregating the predictions of a mixture-of-experts, assuming each expert captures relatively different latent features. The experimental results demonstrate that our post-hoc control over the experts significantly enhances the model's robustness to the distribution shift in shortcuts. Besides, we show that our approach has some practical advantages. We also analyze our model and provide results to support the assumption.