Tag
The paper introduces forget-set misalignment in LLM unlearning and proposes a data-blind framework called CONFS to address it, achieving a competitive forgetting-utility balance.
The paper identifies hidden boundary motion in affine layers of Transformers, where weight updates act as bias updates due to nonzero input mean, and proposes SBO-AdamW to orthogonalize shape and boundary components, yielding improved validation accuracy.
This paper introduces Gradient Fingerprint (GRIFT), a method for detecting reward hacking in reinforcement learning with verifiable rewards by analyzing models' internal gradient computations rather than surface-level reasoning traces. The approach achieves over 25% relative improvement in detecting implicit reward-hacking behaviors across math, code, and logical reasoning benchmarks.