11h ago
Shape of Memory: a Geometric Analysis of Machine Unlearning in Second-Order Optimizers
★★★★★
significance 2/5
The paper investigates how second-order optimizers retain information during machine unlearning tasks. It identifies that while performance may appear to align with ideal deletion, the optimizer state exhibits significant volatility, suggesting residual information remains detectable.
Why it matters
Residual information patterns in optimizer states suggest current unlearning methods may fail to achieve true data erasure at a structural level.
Tags
#machine unlearning #optimizers #optimization #information theoryRelated coverage
- Global South OpportunitiesPivotal Research Fellowship 2026 (Q3): AI Safety Research Opportunity - Global South Opportunities
- arXiv cs.AIAn Intelligent Fault Diagnosis Method for General Aviation Aircraft Based on Multi-Fidelity Digital Twin and FMEA Knowledge Enhancement
- arXiv cs.AIPExA: Parallel Exploration Agent for Complex Text-to-SQL
- arXiv cs.AIThe Power of Power Law: Asymmetry Enables Compositional Reasoning
- arXiv cs.AIOn the Existence of an Inverse Solution for Preference-Based Reductions in Argumentation