Improving Knowledge Distillation Under Unknown Covariate Shift Through Confidence-Guided Data Augmentation
This work addresses the practical issue of unknown spurious features in knowledge distillation for deploying robust small models, representing an incremental advance in data augmentation methods.
The paper tackles the problem of covariate shift in knowledge distillation by introducing a diffusion-based data augmentation strategy that generates challenging samples to maximize teacher-student disagreement, resulting in significant improvements in worst group and mean group accuracy on datasets like CelebA and SpuCo Birds, and spurious mAUC on spurious ImageNet.
Large foundation models trained on extensive datasets demonstrate strong zero-shot capabilities in various domains. To replicate their success when data and model size are constrained, knowledge distillation has become an established tool for transferring knowledge from foundation models to small student networks. However, the effectiveness of distillation is critically limited by the available training data. This work addresses the common practical issue of covariate shift in knowledge distillation, where spurious features appear during training but not at test time. We ask the question: when these spurious features are unknown, yet a robust teacher is available, is it possible for a student to also become robust to them? We address this problem by introducing a novel diffusion-based data augmentation strategy that generates images by maximizing the disagreement between the teacher and the student, effectively creating challenging samples that the student struggles with. Experiments demonstrate that our approach significantly improves worst group and mean group accuracy on CelebA and SpuCo Birds as well as the spurious mAUC on spurious ImageNet under covariate shift, outperforming state-of-the-art diffusion-based data augmentation baselines