An Information Theoretic Evaluation Metric For Strong Unlearning

2024-05-28

AI Summary

Evaluating strong unlearning in deep neural networks—i.e., verifying whether a model after unlearning is statistically indistinguishable from one never exposed to the to-be-forgotten data—remains challenging. Existing black-box methods, relying solely on output-layer behavior, lack sensitivity to residual information persisting in intermediate layers. Method: We propose the first white-box evaluation metric, the Information Difference Index (IDI), grounded in mutual information theory. IDI quantifies the statistical dependence between intermediate-layer representations and the target forget-label, enabling fine-grained detection of lingering memorization. Contribution/Results: Extensive experiments across diverse architectures (ResNet, ViT) and datasets (CIFAR-10/100, ImageNet subsets) demonstrate that IDI achieves superior sensitivity and cross-model consistency compared to black-box baselines—including membership inference attacks and accuracy-difference metrics. IDI provides an interpretable, reproducible, and theoretically principled benchmark for compliance verification under the “right to be forgotten.”

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Adversarial Attacks & Robustness
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Security and Privacy: Large-scale security measurements

Abstract

Machine unlearning (MU) aims to remove the influence of specific data from trained models, addressing privacy concerns and ensuring compliance with regulations such as the ``right to be forgotten.''Evaluating strong unlearning, where the unlearned model is indistinguishable from one retrained without the forgetting data, remains a significant challenge in deep neural networks (DNNs). Common black-box metrics, such as variants of membership inference attacks and accuracy comparisons, primarily assess model outputs but often fail to capture residual information in intermediate layers. To bridge this gap, we introduce the Information Difference Index (IDI), a novel white-box metric inspired by information theory. IDI quantifies retained information in intermediate features by measuring mutual information between those features and the labels to be forgotten, offering a more comprehensive assessment of unlearning efficacy. Our experiments demonstrate that IDI effectively measures the degree of unlearning across various datasets and architectures, providing a reliable tool for evaluating strong unlearning in DNNs.

Problem

• Evaluating strong unlearning remains challenging in deep neural networks
• Existing black-box metrics fail to capture residual information in layers
• Proposing a white-box metric to quantify retained forgotten information

Innovation

• IDI measures mutual information in intermediate layers
• IDI quantifies retained information for forgotten labels
• IDI offers white-box evaluation using information theory