paper-with-me

Papers

Generative Reward Models

2024-10-02 · Dakota Mahan, Duy Van Phung, Rafael Rafailov, Chase Blagden, Nathan Lile, Louis Castricato, Jan-Philipp Fränken, Chelsea Finn, Alon Albalak

Reinforcement Learning from Human Feedback (RLHF) has greatly improved the performance of modern Large Language Models (LLMs). The RLHF process is resource-intensive and technically challenging, generally requiring a large collection of human preference labels over model-generated outputs. Reinforcement Learning from AI Feedback (RLAIF) addresses this data collection challenge by leveraging synthetic preferences generated by an LLM. However, recent work has shown that synthetic preferences labels may not align well with human preference judgments. To address this, we propose a hybrid approach that unifies RLHF and RLAIF methodologies. We introduce GenRM, an iterative algorithm that trains an LLM on self-generated reasoning traces, leading to synthetic preference labels matching human preference judgments. Empirically, we show that zero-shot LLM-based judgments under-perform compared to Bradley-Terry reward models on in-distribution tasks (between 9-36%). In contrast, GenRM achieves in-distribution accuracy comparable to Bradley-Terry models, while significantly outperforming them on out-of-distribution tasks (between 10-45%). Moreover, GenRM surpasses the performance of using LLMs as judges on both in-distribution (by 9-31%) and out-of-distribution tasks (by 2- 6%). Our results show that combining the strengths of RLHF and RLAIF offers a promising approach for improving the quality of synthetic preference labels.

📄 PDF Abstract BibTeX arXiv:2410.12832

Code (0)

등록된 구현이 없습니다.

Tasks

reinforcement-learningReinforcement Learning

Methods 이 논문이 사용한 방법론

ALIGN In the ALIGN method, visual and language representations are jointly trained from noisy image alt-text data. The image and text encoders are learned via contrastive loss…
RLAIF 설명 없음

Similar Papers 제목 키워드 기반

RRC: Unlocking Generative Reward Models in LLM Reinforcement Learning via Ranking-Based Reward Construction

2026-08-06 · Chenglong Wang, Ziming Zhu, Yifu Huo, Bei Li 외 arxiv

Recent advances in reward modeling show a paradigm shift from discriminative reward models to generative reward models. However, despite their strong capabilities in response ranking, generative reward models have not re…

Reinforcement Learning

DiT-Reward: Generative Representations for Text-to-Image Reward Modeling

2026-06-22 · Yuanming Yang, Guoqing Ma, Bo Wang, Yuan Zhang 외 arxiv

Can representations learned for image generation also support the evaluation of generated images? We study text-to-image reward prediction as a downstream task of generative representation learning. To this end, we intro…

Representation LearningImage Generation

Interpretable Reward Redistribution in Reinforcement Learning: A Causal Approach

2023-05-28 · NeurIPS 2023 11 · Yudi Zhang, Yali Du, Biwei Huang, Ziyan Wang 외

A major challenge in reinforcement learning is to determine which state-action pairs are responsible for future rewards that are delayed. Reward redistribution serves as a solution to re-assign credits for each time step…

reinforcement-learningReinforcement Learning

GRAM-R$^2$: Self-Training Generative Foundation Reward Models for Reward Reasoning

2025-09-02 · Chenglong Wang, Yongyu Mu, Hang Zhou, Yifu Huo 외 arxiv

Significant progress in reward modeling over recent years has been driven by a paradigm shift from task-specific designs towards generalist reward models. Despite this trend, developing effective reward models remains a …

Reinforcement Learning

GRAM: A Generative Foundation Reward Model for Reward Generalization

2025-06-17 · Chenglong Wang, Yang Gan, Yifu Huo, Yongyu Mu 외

In aligning large language models (LLMs), reward models have played an important role, but are standardly trained as discriminative models and rely only on labeled human preference data. In this paper, we explore methods…