Understanding and Mitigating Miscalibration in Prompt Tuning for Vision-Language Models

Confidence calibration is critical for the safe deployment of machine learning models in the real world. However, such issue in vision-language models like CLIP, particularly after fine-tuning, has not been fully addressed. In this work, we demonstrate that existing prompt tuning methods usually lead to a trade-off of calibration between base and new classes: the cross-entropy loss used in standard prompt tuning (e.g., CoOp) causes overconfidence in new classes by increasing textual label divergence, whereas regularizationbased tuning (e.g., KgCoOp) maintains the confidence level but results in underconfidence in base classes due to the improved accuracy. Inspired by the observations, we introduce Dynamic Outlier Regularization (DOR) to ensure the confidence calibration on both base and new classes after finetuning. In particular, DOR minimizes the feature deviation of novel textual labels (instead of base classes) sampled from a large vocabulary set. In effect, DOR prevents the increase in textual divergence for new labels while easing restrictions on base classes. Extensive experiments demonstrate that DOR can notably enhance the calibration performance of current fine-tuning methods. Our code is available at https://github.com/ml-stat-Sustech/Outlier-Calibration .