analiza aiqbench
This research introduces Spectral Saliency Unlearning (SSU), a novel method that leverages spectral analysis to efficiently remove specific data influences from AI models while maintaining overall utility across classifiers, diffusion models, and LLMs. By thresholding weak singular components based on theoretical justification for the forgetting-retention trade-off, SSU offers a more targeted approach compared to previous gradient-based variants like Muon. The work advances machine unlearning by providing practical mechanisms to address privacy concerns without sacrificing model performance.
Spectral Saliency for Machine Unlearning
arXiv:2608.15548v1 Announce Type: new Abstract: Machine unlearning (MU) aims to remove the influence of specific training data while preserving model utility. As the name suggests, MU can be viewed as the inverse of learning, using gradient-based updates to reduce the influence of a forget-set by counteracting the previously learned behavior. Recently, Muon, a gradient descent variant, has been introduced. Muon applies spectral magnitude normalization to encourage exploration of rare directions and demonstrates promising performance. Inspired by Muon, we adopt the spectral view for unlearning and propose Spectral Saliency Unlearning (SSU). SSU thresholds weak singular components and updates only those directions supported by a confident unlearning signal. We further provide theoretical justification for this thresholding approach from the perspective of the forgetting-retention trade-off. Experiments across image classifiers, diffusion models, and LLMs demonstrate SSU's effectiveness.
Abstract:Machine unlearning (MU) aims to remove the influence of specific training data while preserving model utility. As the name suggests, MU can be viewed as the inverse of learning, using gradient-based updates to reduce the influence of a forget-set by counteracting the previously learned behavior. Recently, Muon, a gradient descent variant, has been introduced. Muon applies spectral magnitude normalization to encourage exploration of rare directions and demonstrates promising performance. Inspired by Muon, we adopt the spectral view for unlearning and propose Spectral Saliency Unlearning (SSU). SSU thresholds weak singular components and updates only those directions supported by a confident unlearning signal. We further provide theoretical justification for this thresholding approach from the perspective of the forgetting-retention trade-off. Experiments across image classifiers, diffusion models, and LLMs demonstrate SSU's effectiveness.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2608.15548 [cs.LG]
(or arXiv:2608.15548v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2608.15548 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 225 cuvinte
Subiecte conexe
Gradient Concentration, Not Weight Saliency, Explains Representation-Level Class Unlearning
arXiv · 24 iul. 2026
QR-Erase: Efficient Subspace-Based Machine Unlearning with Layer Localization
arXiv · 4 aug. 2026
SPACE: Source-free Proxy Anchor Concept Erasure for MLLMs
arXiv · 10 iun. 2026
State-Space Unlearning for Non-Stationary Bias in Land Surface Forecasting
arXiv · 5 oct. 2026