Causal Reward Adjustment: Mitigating Reward Hacking in External Reasoning via Backdoor Correction
External reasoning systems combine language models with process reward models (PRMs) to select high-quality reasoning paths for complex tasks such as mathematical problem solving. However, these systems are prone to reward hacking, where high-scoring but logically incorrect paths are assigned high scores by the PRMs, leading to incorrect answers. From a causal inference perspective, we attribute this phenomenon primarily to the presence of confounding semantic features. To address it, we propose Causal Reward Adjustment (CRA), a method that mitigates reward hacking by estimating the true reward of a reasoning path. CRA trains sparse autoencoders on the PRM's internal activations to recover interpretable features, then corrects confounding by using backdoor adjustment. Experiments on math solving datasets demonstrate that CRA mitigates reward hacking and improves final accuracy, without modifying the policy model or retraining PRM.
Code (0)
등록된 구현이 없습니다.
Tasks
Causal InferenceSimilar Papers 제목 키워드 기반
Factored Causal Representation Learning for Robust Reward Modeling in RLHF
A reliable reward model is essential for aligning large language models with human preferences through reinforcement learning from human feedback. However, standard reward models are susceptible to spurious features that…
Representation LearningReinforcement LearningThe Energy Loss Phenomenon in RLHF: A New Perspective on Mitigating Reward Hacking
This work identifies the Energy Loss Phenomenon in Reinforcement Learning from Human Feedback (RLHF) and its connection to reward hacking. Specifically, energy loss in the final layer of a Large Language Model (LLM) grad…
Large Language ModelProof-of-Use: Mitigating Tool-Call Hacking in Deep Research Agents
While reinforcement learning (RL) enhances their ability to plan and reason across retrieval steps, we identify a critical failure mode in this setting: Tool-Call Hacking. Unlike execution-based tools (e.g., code or math…
Reinforcement LearningThe Horcrux: Mechanistically Interpretable Task Decomposition for Detecting and Mitigating Reward Hacking in Embodied AI Systems
Embodied AI agents exploit reward signal flaws through reward hacking, achieving high proxy scores while failing true objectives. We introduce Mechanistically Interpretable Task Decomposition (MITD), a hierarchical trans…
Sail into the Headwind: Alignment via Robust Rewards and Dynamic Labels against Reward Hacking
Aligning AI systems with human preferences typically suffers from the infamous reward hacking problem, where optimization of an imperfect reward model leads to undesired behaviors. In this paper, we investigate reward ha…
Mathematical Reasoning