Erased, but Not Gone: Output Forgetting Is Not True Forgetting

2026-06-23

AI Summary

This work addresses a critical limitation in existing machine unlearning methods, which predominantly rely on output-layer metrics to assess forgetting efficacy, often failing to capture whether the model truly achieves representation-level equivalence to a model trained from scratch without the forgotten data. To bridge this gap, the paper proposes a novel evaluation paradigm—“retraining-consistent representation unlearning”—using a retrained model as the gold standard. By integrating representation space alignment, directional residual analysis, and asymmetry detection between forgotten and retained samples, the study systematically demonstrates that mainstream unlearning algorithms, while appearing effective at the output level, retain significant structural biases in their internal representations. This finding challenges conventional evaluation practices that depend solely on output metrics and reveals a systematic overestimation of actual unlearning performance in current approaches.

Abstract

Machine unlearning (MU) is commonly judged by output forgetting, such as low forget-set accuracy or reduced logit-level membership inference. But if output-level success can coexist with retraining-inconsistent residuals in representation space, what kind of forgetting are current evaluations actually certifying? We study this question through retraining-consistent representation forgetting, using the retrained model (i.e., trained from scratch without the forget data) as an operational reference for correct forgetting. Across multiple unlearning methods, datasets, and models, our theoretical analysis and empirical results show that standard output-level evaluation can systematically overestimate the success of unlearning. Under this stronger lens, current methods often appear forgotten at the output layer while exhibiting a structured mismatch relative to retraining. They partially align with retraining on forget samples, remain more inconsistent on retain samples, and leave residual discrepancy concentrated along retraining-related directions rather than diffuse in representation space. This structured mismatch is characterized by forget/retain asymmetry, directional mismatch, and concentrated residuals along retraining-related directions. These results suggest that current MU is often evaluated for apparent forgetting rather than retraining-consistent forgetting. More broadly, retraining reveals what output forgetting hides.

Problem

• machine unlearning
• output forgetting
• retraining-consistent forgetting
• representation space
• evaluation bias

Innovation

• machine unlearning
• representation forgetting
• retraining consistency
• structured residuals
• output forgetting