H2DF: Calibrated Target Bands for Controllable Language-Model Unlearning

Not provided in the abstract

unlearninglanguage modelingforgetting controlcalibrated objectives

2610.26864

Builder Relevance

70%

2h ago

Abstract

H2DF introduces a calibrated target-band objective to improve controllable language-model unlearning by managing forget loss effectively.

Reality Card

Core Claim

H2DF achieves a forget NLL of 1.528 and a retain NLL of 1.165 while using retain replay only every tenth step, demonstrating lower variability than gradient ascent.

Method / Result

Forget NLL of 1.528 with margin 0.049 and retain NLL of 1.165 with margin 0.047.

Limitations

Sensitivity to random seed and calibration quality may affect reproducibility.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.

← Back to all papers