English
Related papers

Related papers: Scaling Laws for Reward Model Overoptimization

200 papers

In reinforcement learning (RL), agents continually interact with the environment and use the feedback to refine their behavior. To guide policy optimization, reward models are introduced as proxies of the desired objectives, such that when…

Machine Learning · Computer Science 2025-06-19 Rui Yu , Shenghua Wan , Yucen Wang , Chen-Xiao Gao , Le Gan , Zongzhang Zhang , De-Chuan Zhan

Policy alignment of large language models refers to constrained policy optimization, where the policy is optimized to maximize a reward while staying close to a reference policy with respect to an $f$-divergence such as the $\mathsf{KL}$…

Machine Learning · Computer Science 2024-06-11 Youssef Mroueh

Standard reinforcement learning from human feedback (RLHF) trains a reward model on pairwise preference data and then uses it for policy optimization. However, while reward models are optimized to capture relative preferences, existing…

Machine Learning · Computer Science 2026-02-05 Kyuseong Choi , Dwaipayan Saha , Woojeong Kim , Anish Agarwal , Raaz Dwivedi

Preference learning is critical for aligning large language models (LLMs) with human values, with the quality of preference datasets playing a crucial role in this process. While existing metrics primarily assess data quality based on…

Machine Learning · Computer Science 2025-03-05 Kexin Huang , Junkang Wu , Ziqian Chen , Xue Wang , Jinyang Gao , Bolin Ding , Jiancan Wu , Xiangnan He , Xiang Wang

Reward models are widely used as proxies for human preferences when aligning or evaluating LLMs. However, reward models are black boxes, and it is often unclear what, exactly, they are actually rewarding. In this paper we develop…

Computation and Language · Computer Science 2025-05-21 David Reber , Sean Richardson , Todd Nief , Cristina Garbacea , Victor Veitch

Techniques that learn improved representations via offline data or self-supervised objectives have shown impressive results in traditional reinforcement learning (RL). Nevertheless, it is unclear how improved representation learning can…

Computation and Language · Computer Science 2024-10-25 Vaskar Nath , Dylan Slack , Jeff Da , Yuntao Ma , Hugh Zhang , Spencer Whitehead , Sean Hendryx

AI systems often rely on two key components: a specified goal or reward function and an optimization algorithm to compute the optimal behavior for that goal. This approach is intended to provide value for a principal: the user on whose…

Artificial Intelligence · Computer Science 2021-02-09 Simon Zhuang , Dylan Hadfield-Menell

Modern large language models (LLMs) are optimized for human-aligned responses using Reinforcement Learning from Human Feedback (RLHF). However, existing RLHF approaches assume a universal preference model and fail to account for individual…

Machine Learning · Computer Science 2025-03-11 Idan Shenfeld , Felix Faltings , Pulkit Agrawal , Aldo Pacchiano

Reinforcement Learning from Human Feedback (\textbf{RLHF}) has emerged as a dominant approach for aligning LLM outputs with human preferences. Inspired by the success of RLHF, we study the performance of multiple algorithms that learn from…

There are several distinct failure modes for overoptimization of systems on the basis of metrics. This occurs when a metric which can be used to improve a system is used to an extent that further optimization is ineffective or harmful, and…

Artificial Intelligence · Computer Science 2019-02-26 David Manheim , Scott Garrabrant

To align conditional text generation model outputs with desired behaviors, there has been an increasing focus on training the model using reinforcement learning (RL) with reward functions learned from human annotations. Under this…

Computation and Language · Computer Science 2023-06-02 Richard Yuanzhe Pang , Vishakh Padmakumar , Thibault Sellam , Ankur P. Parikh , He He

Collecting high-quality preference datasets for reinforcement learning from human feedback (RLHF) is resource-intensive and challenging. As a result, researchers often train reward models on extensive offline datasets which aggregate…

Machine Learning · Computer Science 2024-12-17 Shambhavi Krishna , Aishwarya Sahoo

Reinforcement Learning from Human Feedback (RLHF) is a powerful paradigm for aligning foundation models to human values and preferences. However, current RLHF techniques cannot account for the naturally occurring differences in individual…

Machine Learning · Computer Science 2024-08-20 Sriyash Poddar , Yanming Wan , Hamish Ivison , Abhishek Gupta , Natasha Jaques

Reinforcement Learning from Human Feedback (RLHF) is the standard method to align Large Language Models (LLMs) with human preferences. In this work, we introduce alignment tampering, a potential vulnerability where the LLM undergoing…

Artificial Intelligence · Computer Science 2026-05-27 Dongyoon Hahm , Dylan Hadfield-Menell , Kimin Lee

Designing robust reinforcement learning (RL) agents in the presence of imperfect reward signals remains a core challenge. In practice, agents are often trained with proxy rewards that only approximate the true objective, leaving them…

Machine Learning · Computer Science 2026-04-15 Zixuan Liu , Xiaolin Sun , Zizhan Zheng

The utility of reinforcement learning is limited by the alignment of reward functions with the interests of human stakeholders. One promising method for alignment is to learn the reward function from human-generated preferences between…

Machine Learning · Computer Science 2023-09-08 W. Bradley Knox , Stephane Hatgis-Kessell , Serena Booth , Scott Niekum , Peter Stone , Alessandro Allievi

We study estimation and statistical inference for reward models used in aligning large language models (LLMs). A key component of LLM alignment is reinforcement learning from human feedback (RLHF), where humans compare pairs of…

Machine Learning · Statistics 2025-12-04 Pangpang Liu , Junwei Lu , Will Wei Sun

Generative foundation models are susceptible to implicit biases that can arise from extensive unsupervised training data. Such biases can produce suboptimal samples, skewed outcomes, and unfairness, with potentially serious consequences.…

Machine Learning · Computer Science 2023-12-04 Hanze Dong , Wei Xiong , Deepanshu Goyal , Yihan Zhang , Winnie Chow , Rui Pan , Shizhe Diao , Jipeng Zhang , Kashun Shum , Tong Zhang

Alignment of large language models (LLMs) typically involves training a reward model on preference data, followed by policy optimization with respect to the reward model. However, optimizing policies with respect to a single reward model…

Machine Learning · Computer Science 2025-07-23 Debangshu Banerjee , Kintan Saha , Aditya Gopalan

Compositional generalization refers to correctly interpret novel combinations of known primitives, which remains a major challenge. Existing approaches often rely on supervised fine-tuning, which encourages models to imitate target outputs.…

Machine Learning · Computer Science 2026-05-07 Xiyan Fu , Wei Liu
‹ Prev 1 3 4 5 6 7 10 Next ›