Calibration Meets Explanation: A Simple and Effective Approach for Model Confidence Estimates

TL;DR

This paper introduces CME, a method that uses model explanations to improve confidence calibration in language models, demonstrating consistent performance gains across multiple datasets and settings.

Contribution

The paper presents a novel explanation-based calibration method, CME, that enhances confidence estimates by leveraging feature attributions, a concept not previously explored for this purpose.

Findings

CME improves calibration across six datasets and two language models.

Combining CME with temperature scaling further reduces calibration errors.

Model explanations can effectively aid in confidence calibration.

Abstract

Calibration strengthens the trustworthiness of black-box models by producing better accurate confidence estimates on given examples. However, little is known about if model explanations can help confidence calibration. Intuitively, humans look at important features attributions and decide whether the model is trustworthy. Similarly, the explanations can tell us when the model may or may not know. Inspired by this, we propose a method named CME that leverages model explanations to make the model less confident with non-inductive attributions. The idea is that when the model is not highly confident, it is difficult to identify strong indications of any class, and the tokens accordingly do not have high attribution scores for any class and vice versa. We conduct extensive experiments on six datasets with two popular pre-trained language models in the in-domain and out-of-domain settings. The results show that CME improves calibration performance in all settings. The expected calibration errors are further reduced when combined with temperature scaling. Our findings highlight that model explanations can help calibrate posterior estimates.