Redirection for Erasing Memory (REM): Towards a universal unlearning method for corrupted data

2025-05-23

AI Summary

This work addresses the general machine unlearning problem for vision classifiers under diverse data corruptions—including label noise, backdoor attacks, and semantic contamination—by proposing the first cross-corruption comparable two-dimensional conceptual space (discovery rate × statistical regularity) to unify unlearning task modeling. Methodologically, we introduce a novel “redirect–erase” paradigm: structural expansion incorporates dedicated neurons to dynamically capture corruption-specific responses, followed by precise erasure via freezing or pruning; this is complemented by corruption-response isolation and lightweight parameter adaptation. Evaluated across the full spectrum of corruption types, our approach significantly outperforms state-of-the-art methods, achieving an average 27.4% improvement in unlearning success rate and a 41.8% reduction in generalization error. To our knowledge, this is the first framework enabling universal, robust, and interpretable unlearning for vision models.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Security and Privacy: Security and privacy of machine learning and AI applications
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Machine unlearning is studied for a multitude of tasks, but specialization of unlearning methods to particular tasks has made their systematic comparison challenging. To address this issue, we propose a conceptual space to characterize diverse corrupted data unlearning tasks in vision classifiers. This space is described by two dimensions, the discovery rate (the fraction of the corrupted data that are known at unlearning time) and the statistical regularity of the corrupted data (from random exemplars to shared concepts). Methods proposed previously have been targeted at portions of this space and-we show-fail predictably outside these regions. We propose a novel method, Redirection for Erasing Memory (REM), whose key feature is that corrupted data are redirected to dedicated neurons introduced at unlearning time and then discarded or deactivated to suppress the influence of corrupted data. REM performs strongly across the space of tasks, in contrast to prior SOTA methods that fail outside the regions for which they were designed.

Problem

• Develops a universal unlearning method for corrupted data in vision classifiers
• Addresses task specialization challenges in comparing unlearning methods
• Introduces REM to redirect and suppress corrupted data influence universally

Innovation

• Introduces REM for universal unlearning
• Redirects corrupted data to dedicated neurons
• Deactivates neurons to suppress corrupted influence