Uncertainty-Penalized Direct Preference Optimization
원문 초록
Aligning Large Language Models (LLMs) to human preferences in content, style, and presentation is challenging, in part because preferences are varied, context-dependent, and sometimes inherently ambiguous. While successful, Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO) are prone to the issue of proxy reward overoptimization. Analysis of the DPO loss reveals a critical need for regularization for mislabeled or ambiguous preference pairs to avoid reward hacking. In this work, we develop a pessimistic framework for DPO by introducing preference uncertainty penalization schemes, inspired by offline reinforcement learning. The penalization serves as a correction to the loss which attenuates the loss gradient for uncertain samples. Evaluation of the methods is performed with GPT2 Medium on the Anthropic-HH dataset using a model ensemble to obtain uncertainty estimates, and shows improved overall performance compared to vanilla DPO, as well as better completions on prompts from high-uncertainty chosen/rejected responses.
함께 살펴볼 논문
공통 태그를 근거로 연결했습니다. 실제 인용 관계나 논문의 우수성을 뜻하지 않습니다.
Reasoning on a Budget: Miniaturizing DeepSeek R1 with SFT-GRPO Alignment for Instruction-Tuned LLMs
Meta Reinforcement Learning for Multi-Task Offloading in Vehicular Edge Computing
Investigating the Performance and Reliability, of the Q-Learning Algorithm in Various Unknown Environments
Margin Trader: A Reinforcement Learning Framework for Portfolio Management with Margin and Constraints
Catch: Collaborative Feature Set Search for Automated Feature Engineering
Solving Dynamic Traveling Salesman Problems With Deep Reinforcement Learning
토론
이 논문과 연결된 포럼 글이 아직 없습니다.