SCOPE: Entanglement Frontier Escape for Source-Free Class Unlearning
2026-08-03 • Machine Learning
Machine Learning
AI summaryⓘ
The authors study how to erase an entire class of data from a trained model without access to the original training data, a process called source-free class unlearning. They find that using one fixed method to remove features related to the forget class also accidentally affects features of the classes meant to be kept due to overlapping representations. They prove this issue is a fundamental trade-off and show that conditioning the erasure on each input can avoid it. They propose a new method called SCOPE that adjusts removal based on the model's own predictions, which is efficient and outperforms previous methods across multiple tasks and model types.
source-free unlearningclass unlearningfeature spacerepresentation learningprojectionspectral methodsconditional erasureforget setretain datamodel retraining
Authors
Junhao Cai, Dohun Kim, Sung Il Choi, Juhyun Park, Chengjun Jin, Dowon Kim, Changhee Joo
Abstract
Source-free class unlearning erases whole classes using only the forget data, judged at the representation level, where features can leak a class the head no longer predicts. Existing feature-space erasers answer with one fixed projection, yet forget and retain classes share a representation, so deleting one disturbs the other where they overlap. We prove this tension is a frontier. Every fixed projection that deletes pays a retain cost of at least the retain-readout energy along the forget-discriminant subspace, and erasing that subspace alone attains the floor. The leading source-free erasers all instantiate the form it binds, so the frontier limits the whole class. Conditioning the erasure on the input escapes it. Spectral Conditional Projective Erasure (SCOPE) does so with a single gate, suppressing the forget subspace chiefly on inputs its frozen head's weight scores read as a forget class. It is closed form, needs no retain data or gradient training, and costs orders of magnitude less than retraining. Across five object, face, and speaker benchmarks spanning two modalities and both convolutional and transformer backbones, the frontier predicts the measured retain cost. SCOPE leads the source-free erasers on every benchmark and forget-set size, and at the hardest setting it tops every unlearner, trained methods included.