SOURCE-LINKED INTELLIGENCE

The BatchNorm Illusion: Diagnosing Normalization Artifacts in Machine Unlearning Evaluation

arXiv · AI, language, vision and robotics · article · Sep 8, 2026 · UTC

Approximate machine unlearning aims to remove the influence of specific training data from a trained model without retraining from scratch. We identify a previously undocumented confound in how unlearning is evaluated on BatchNorm-based architectures: a single forward pass over retain data, an operation that modifies no weight, can deterministically rewrite the model's normalization state and reverse the apparent surface-metric forgetting. We formalize this operation as a weight-preserving fixed-point operator and prove that any pre-versus-post gap it induces is provably attributable to BN run

Read original source ↗ Open in workspace

recordType

paper

region

Global