Upcycled and Merged MoE Reward Model for Mitigating Reward Hacking

• URL: http://arxiv.org/abs/2512.00724v1
• Date: Sun, 30 Nov 2025 04:36:37 GMT
• Title: Upcycled and Merged MoE Reward Model for Mitigating Reward Hacking
• Authors: Lingling Fu,
• Abstract summary: We propose an upcycle and merge MoE reward modeling approach to improve robustness and efficiency of RLHF reward models.<n>Our work highlights the potential of upcycle and merge MoE structures for improving both robustness and efficiency of RLHF reward models.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Reward models play a critical role in Reinforcement Learning from Human Feedback (RLHF) by assessing the consistency between generated outputs and human preferences. However, conventional reward models are prone to reward hacking or over-optimization, where the policy exploits shortcut patterns to obtain high reward scores that do not reflect true human preference. Although Mixture-of-Experts (MoE)-based reward models can enhance discriminative capability, they typically introduce substantial computational overhead. To address these challenges, we propose an upcycle and merge MoE reward modeling approach. We first upcycle a dense reward model into a MoE architecture, where a shared expert captures general knowledge, while normal experts specialize in instruction-specific patterns. We then apply routing-weight normalization and merge experts back into a dense model through a learnable weight-averaging mechanism, preserving performance gains while significantly reducing inference cost. Experimental results demonstrate that our method effectively mitigates reward hacking across various model scales. Our work highlights the potential of upcycle and merge MoE structures for improving both robustness and efficiency of RLHF reward models.

Related papers

• Mitigating Reward Hacking in RLHF via Bayesian Non-negative Reward Modeling [49.41422138354821]
  We propose a principled reward modeling framework that integrates non-negative factor analysis into the Bradley-Terry preference model.<n>BNRM represents rewards through a sparse, non-negative latent factor generative process.<n>We show that BNRM substantially mitigates reward over-optimization, improves robustness under distribution shifts, and yields more interpretable reward decompositions than strong baselines.
  arXiv  Detail & Related papers  (2026-02-11T08:14:11Z)
• Cooper: Co-Optimizing Policy and Reward Models in Reinforcement Learning for Large Language Models [28.542061921495353]
  There are two mainstream reward paradigms: model-based rewards and rule-based rewards.<n>Both approaches suffer from limitations: rule-based rewards lack robustness, while model-based rewards are vulnerable to reward hacking.<n>We propose Cooper, a RL framework that jointly optimize both the policy model and the reward model.<n>Our experiments show that Cooper not only alleviates reward hacking but also improves end-to-end RL performance, for instance, achieving a 0.54% gain in average accuracy on Qwen2.5-1.5B-Instruct.
  arXiv  Detail & Related papers  (2025-08-07T17:53:56Z)
• Reward Models Can Improve Themselves: Reward-Guided Adversarial Failure Mode Discovery for Robust Reward Modeling [27.11560841914813]
  We introduce REFORM, a self-improving reward modeling framework that enhances robustness by using the reward model itself to guide the generation of falsely scored responses.<n>We evaluate REFORM on two widely used preference datasets Anthropic Helpful Harmless (HH) and PKU Beavertails.
  arXiv  Detail & Related papers  (2025-07-08T21:56:33Z)
• Activation Reward Models for Few-Shot Model Alignment [77.37511364793515]
  We introduce Activation Reward Models (Activation RMs)<n>Activation RMs leverage activation steering to construct well-aligned reward signals using minimal supervision and no additional model finetuning.<n>We demonstrate the effectiveness of Activation RMs in mitigating reward hacking behaviors, highlighting their utility for safety-critical applications.
  arXiv  Detail & Related papers  (2025-07-02T05:10:29Z)
• RED: Unleashing Token-Level Rewards from Holistic Feedback via Reward Redistribution [50.171320156632866]
  Reinforcement learning from human feedback offers a promising approach to aligning large language models with human preferences.<n>Current reward models operate as sequence-to-one models, allocating a single, sparse, and delayed reward to an entire output sequence.<n>We propose a more fine-grained, token-level guidance approach for RL training.
  arXiv  Detail & Related papers  (2024-11-13T02:45:21Z)
• HAF-RM: A Hybrid Alignment Framework for Reward Model Training [51.59246299566669]
  We propose a hybrid alignment framework HaF-RM for reward model training.<n>It offers a principled and effective approach to enhancing the performance and alignment of reward models.
  arXiv  Detail & Related papers  (2024-07-04T23:26:56Z)
• DogeRM: Equipping Reward Models with Domain Knowledge through Model Merging [65.41765072566287]
  We propose textbfDomain knowledtextbfge merged textbfReward textbfModel (DogeRM), a novel framework that integrates domain-specific knowledge into a general reward model by model merging.
  arXiv  Detail & Related papers  (2024-07-01T17:01:54Z)
• RewardBench: Evaluating Reward Models for Language Modeling [100.28366840977966]
  We present RewardBench, a benchmark dataset and code-base for evaluation of reward models. The dataset is a collection of prompt-chosen-rejected trios spanning chat, reasoning, and safety. On the RewardBench leaderboard, we evaluate reward models trained with a variety of methods.
  arXiv  Detail & Related papers  (2024-03-20T17:49:54Z)
• Secrets of RLHF in Large Language Models Part II: Reward Modeling [134.97964938009588]
  We introduce a series of novel methods to mitigate the influence of incorrect and ambiguous preferences in the dataset. We also introduce contrastive learning to enhance the ability of reward models to distinguish between chosen and rejected responses.
  arXiv  Detail & Related papers  (2024-01-11T17:56:59Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.