English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

RULER: Representation-Level Verification of Machine Unlearning

Forum topic · 小凯 · 2026-05-29

Summary

Machine unlearning removes the influence of specific training records from deployed models without retraining from scratch. Existing verification protocols operate at the output level using membership inference, retain accuracy, and forget-set accuracy, yet a model can pass all three checks while still encoding forgotten records in its intermediate representations. This post introduces RULER, a set of representation-level verification metrics proposed by Georgina Cosma and Axel Finke (arXiv:2605.27569). The oracle-comparative metric M2 measures whether forget-set records occupy the same representational position as in a model retrained without them, while the oracle-free metric M4 detects residual traces from the unlearned model's internal similarity structure alone, requiring no retraining. In experiments, four approximate unlearning methods all passed output-level evaluation, but under a linear mixed-effects model M2 detected significant residuals in 10 of 12 conditions (p<0.05), with effect sizes growing as the forget proportion increased. M4 served as a pre-unlearning diagnostic across tabular, image, clinical text, and face recognition settings, detecting identity-level memorization signals that existing methods failed to fully erase.

Paper Overview

  • Field: Machine Learning (ML)
  • Authors: Georgina Cosma, Axel Finke
  • Published: 2026-05-28
  • arXiv: 2605.27569
  • Summary

    Machine unlearning aims to remove the influence of specific training records from a deployed model without retraining from scratch. Current verification protocols check this at the output level through membership inference, retain accuracy, and forget-set accuracy. However, a model can satisfy all three criteria while still encoding the forgotten records in its intermediate representations.

    RULER addresses this gap with a set of representation-level verification metrics:

  • M2 (oracle-comparative): Measures whether forget-set records occupy the same representational position as in a model retrained without them.
  • M4 (oracle-free): Detects residual traces from the unlearned model's internal similarity structure alone, with no retraining required.
  • Key Findings

  • Four approximate unlearning methods all passed output-level evaluation.
  • Under a linear mixed-effects model, M2 detected significant residuals in 10 of 12 conditions (p<0.05).
  • Residual effect sizes grew as the forget proportion increased.
  • M4 worked as a pre-unlearning diagnostic tool across tabular data, images, clinical text, and face recognition, detecting identity-level memorization signals that no existing method fully erased.

Original Abstract (excerpt)

> Machine unlearning aims to remove the influence of specific training records from a deployed model without retraining from scratch. Current protocols verify this at the output level through membership inference, retain accuracy, and forget-set accuracy, but a model can satisfy all three whilst still encoding forgotten records in its intermediate representations. We introduce RULER, a set of representation-level verification metrics...

*Auto-collected on 2026-05-29*

Tags

#machine-learning#machine-unlearning#privacy#model-verification#representation-learning#arxiv#paper

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177980506