SignCert-PO mitigates reward hacking in RLHF by down-weighting completions whose advantage signs are not robust to small reward-model perturbations, using a certified preservation radius derived at the policy optimization stage.
However, under the linear head model, for anyw ′ ∈R d, K ∑ j=1 Aj(w′) = (w ′)T K ∑ j=1 (hj − ¯h) =0, since ∑K j=1(hj − ¯h) = 0 by definition of ¯h
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Mitigating Reward Hacking in RLHF via Advantage Sign Robustness
SignCert-PO mitigates reward hacking in RLHF by down-weighting completions whose advantage signs are not robust to small reward-model perturbations, using a certified preservation radius derived at the policy optimization stage.