Class unlearning methods leak membership through neighbor-class output probabilities, and a tilted reweighting objective that mimics retrained models reduces this leakage.
To evaluate unlearning on a more challenging benchmark, we also apply our method to theTiny-ImageNet- 200dataset using a ResNet-18 backbone
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
On the Necessity of Output Distribution Reweighting for Effective Class Unlearning
Class unlearning methods leak membership through neighbor-class output probabilities, and a tilted reweighting objective that mimics retrained models reduces this leakage.