Regularizing Attention Scores with Bootstrapping

2026-04-01

AI Summary

This work addresses the issue of non-zero noise corrupting attention scores in Vision Transformers, which often leads to blurry and poorly interpretable attention maps. To mitigate this, the authors introduce bootstrapping into the attention mechanism for the first time, leveraging resampled input features to construct a guiding distribution that estimates the significance and posterior probability of attention scores. This approach effectively suppresses spurious attention induced by noise while providing a principled measure of uncertainty and a form of statistical regularization. Evaluated on both natural and medical images, the method consistently enhances the sparsity and focus of attention maps, with compelling performance validated through extensive simulations and real-world experiments.

Technology Category

Computer Vision: Diffusion Models for VisionMachine Learning: Calibration & Uncertainty QuantificationNatural Language Processing: Safety and Robustness

Application Category

Security and Privacy: Data transparency and provenanceSearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingResponsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Vision transformers (ViT) rely on attention mechanism to weigh input features, and therefore attention scores have naturally been considered as explanations for its decision-making process. However, attention scores are almost always non-zero, resulting in noisy and diffused attention maps and limiting interpretability. Can we quantify uncertainty measures of attention scores and obtain regularized attention scores? To this end, we consider attention scores of ViT in a statistical framework where independent noise would lead to insignificant yet non-zero scores. Leveraging statistical learning techniques, we introduce the bootstrapping for attention scores which generates a baseline distribution of attention scores by resampling input features. Such a bootstrap distribution is then used to estimate significances and posterior probabilities of attention scores. In natural and medical images, the proposed \emph{Attention Regularization} approach demonstrates a straightforward removal of spurious attention arising from noise, drastically improving shrinkage and sparsity. Quantitative evaluations are conducted using both simulation and real-world datasets. Our study highlights bootstrapping as a practical regularization tool when using attention scores as explanations for ViT. Code available: https://github.com/ncchung/AttentionRegularization

Problem

• attention scores
• interpretability
• vision transformers
• regularization
• uncertainty quantification

Innovation

• Attention Regularization
• Bootstrapping
• Vision Transformers
• Interpretability
• Statistical Significance