CoChat
Sign Up
PreprintCited by 4Reinforcement Learning in RoboticsNatural Language Processing TechniquesTopic Modeling

The DOI matches its DataCite record; 2 of the 2 references it cites were found in a registry or an index.

Pairwise Proximal Policy Optimization: Harnessing Relative Feedback for LLM Alignment

Open access
Wu, Tianhao, Zhu, Banghua, Zhang, Ruoyu, Wen, Zhaojin, Ramchandran, Kannan, Jiao, Jiantao·2023·arXiv
DOI 10.48550/arxiv.2310.00212arXiv 2310.00212
Open this paper in CoChat — free accountOpen PDF

Abstract

Large Language Models (LLMs) can acquire extensive world knowledge through pre-training on large corpora. However, due to exposure to low-quality data, LLMs may exhibit harmful behavior without aligning with human values. The dominant approach for steering LLMs towards beneficial behavior involves Reinforcement Learning with Human Feedback (RLHF), with Proximal Policy Optimization (PPO) serving as the default RL optimizer. Despite its effectiveness, PPO has limitations when optimizing rewards trained from comparison-based loss. Primarily, PPO is not invariant to equivalent reward functions containing identical preference information due to the need to calibrate the reward scale. Additionally, PPO's necessity for token-wise updates introduces complexity in both function approximation and algorithm design compared to trajectory-wise optimization. This paper proposes a new framework, reinforcement learning with relative feedback, and a novel trajectory-wise policy gradient algorithm, Pairwise Proximal Policy Optimization (P3O) that operates directly on comparative rewards. We show theoretically that P3O is invariant to equivalent rewards and avoids the complexity of PPO. Empirical evaluations demonstrate that P3O outperforms PPO in the KL-Reward trade-off and can align with human preferences as well as or better than prior methods. In summary, this work introduces a simpler yet effective approach for aligning LLMs to human preferences through relative feedback.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

DataCite · 10.48550/arxiv.2310.00212

Cites

2 of 2 references verified0 citation errors · 0 not checkable

Statistics

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • RLHF-PPO has shown promising results but suffers from sensitivity to hyperparameters, the need for training additional models, and potential misalignment of the RM.

    “While RLHF-PPO has shown promising results, it suffers from sensitivity to hyperparameters, the need for training additional models, and potential misalignment of the RM [138,149,174,197,220].”

    From RLHF Deciphered: A Critical Analysis of Reinforcement Learning from Human Feedback for LLMs (2026) · cites this paper as [197]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Flagged first, then by citations

Research works printed but not linked: 42

  1. Yuntao Bai, Andy Jones, Kamal Ndousse, Amanda Askell, Anna Chen, Nova DasSarma, Dawn Drain, Stanislav Fort, Deep Ganguli, Tom Henighan, et al. Training a helpful and harmless assistant with reinforcement learning from human feedback. arXiv preprint arXiv:2204.05862, 2022a. · arXiv
  2. Yuntao Bai, Saurav Kadavath, Sandipan Kundu, Amanda Askell, Jackson Kernion, Andy Jones, Anna Chen, Anna Goldie, Azalia Mirhoseini, Cameron McKinnon, et al. Constitutional ai: Harm- lessness from ai feedback. arXiv preprint arXiv:2212.08073, 2022b. · arXiv
  3. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. Advances in neural information processing systems, 33:1877-1901, 2020.
  4. Sébastien Bubeck, Varun Chandrasekaran, Ronen Eldan, Johannes Gehrke, Eric Horvitz, Ece Ka- mar, Peter Lee, Yin Tat Lee, Yuanzhi Li, Scott Lundberg, et al. Sparks of artificial general intelligence: Early experiments with gpt-4. arXiv preprint arXiv:2303.12712, 2023. · arXiv
  5. Yi Chen, Rui Wang, Haiyun Jiang, Shuming Shi, and Ruifeng Xu. Exploring the use of large language models for reference-free text quality evaluation: A preliminary empirical study. arXiv preprint arXiv:2304.00723, 2023. · arXiv
  6. Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, et al. Palm: Scaling language modeling with pathways. arXiv preprint arXiv:2204.02311, 2022. · arXiv
  7. Paul F Christiano, Jan Leike, Tom Brown, Miljan Martic, Shane Legg, and Dario Amodei. Deep reinforcement learning from human preferences. Advances in neural information processing sys- tems, 30, 2017.
  8. Hanze Dong, Wei Xiong, Deepanshu Goyal, Rui Pan, Shizhe Diao, Jipeng Zhang, Kashun Shum, and Tong Zhang. Raft: Reward ranked finetuning for generative foundation model alignment. arXiv preprint arXiv:2304.06767, 2023. · arXiv
  9. Miroslav Dudík, Katja Hofmann, Robert E Schapire, Aleksandrs Slivkins, and Masrour Zoghi. Con- textual dueling bandits. In Conference on Learning Theory, pp. 563-587. PMLR, 2015.
  10. Logan Engstrom, Andrew Ilyas, Shibani Santurkar, Dimitris Tsipras, Firdaus Janoos, Larry Rudolph, and Aleksander Madry. Implementation matters in deep policy gradients: A case study on ppo and trpo. arXiv preprint arXiv:2005.12729, 2020. · arXiv
  11. Deep Ganguli, Liane Lovitt, Jackson Kernion, Amanda Askell, Yuntao Bai, Saurav Kadavath, Ben Mann, Ethan Perez, Nicholas Schiefer, Kamal Ndousse, et al. Red teaming language models to reduce harms: Methods, scaling behaviors, and lessons learned. arXiv preprint arXiv:2209.07858, 2022. · arXiv
  12. Leo Gao, John Schulman, and Jacob Hilton. Scaling laws for reward model overoptimization. In International Conference on Machine Learning, pp. 10835-10866. PMLR, 2023.
  13. Caglar Gulcehre, Tom Le Paine, Srivatsan Srinivasan, Ksenia Konyushkova, Lotte Weerts, Abhishek Sharma, Aditya Siddhant, Alex Ahern, Miaosen Wang, Chenjie Gu, et al. Reinforced self-training (rest) for language modeling. arXiv preprint arXiv:2308.08998, 2023. · arXiv
  14. Ashesh Jain, Brian Wojcik, Thorsten Joachims, and Ashutosh Saxena. Learning trajectory pref- erences for manipulators via iterative improvement. Advances in neural information processing systems, 26, 2013.
  15. Andras Kupcsik, David Hsu, and Wee Sun Lee. Learning dynamic robot-to-human object handover from human feedback. Robotics Research: Volume 1, pp. 161-176, 2018.
  16. Tor Lattimore and Csaba Szepesvári. Bandit algorithms. Cambridge University Press, 2020.
  17. Harrison Lee, Samrat Phatale, Hassan Mansoor, Kellie Lu, Thomas Mesnard, Colton Bishop, Victor Carbune, and Abhinav Rastogi. Rlaif: Scaling reinforcement learning from human feedback with ai feedback. arXiv preprint arXiv:2309.00267, 2023. · arXiv
  18. Hunter Lightman, Vineet Kosaraju, Yura Burda, Harri Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. Let's verify step by step. arXiv preprint arXiv:2305.20050, 2023. · arXiv
  19. Volodymyr Mnih, Adria Puigdomenech Badia, Mehdi Mirza, Alex Graves, Timothy Lillicrap, Tim Harley, David Silver, and Koray Kavukcuoglu. Asynchronous methods for deep reinforcement learning. In International conference on machine learning, pp. 1928-1937. PMLR, 2016.
  20. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback. Advances in Neural Information Processing Systems, 35: 27730-27744, 2022.
  21. Ethan Perez, Saffron Huang, Francis Song, Trevor Cai, Roman Ring, John Aslanides, Amelia Glaese, Nat McAleese, and Geoffrey Irving. Red teaming language models with language models. arXiv preprint arXiv:2202.03286, 2022. · arXiv
  22. Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D Manning, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. arXiv preprint arXiv:2305.18290, 2023. · arXiv
  23. Rajkumar Ramamurthy, Prithviraj Ammanabrolu, Kianté Brantley, Jack Hessel, Rafet Sifa, Chris- tian Bauckhage, Hannaneh Hajishirzi, and Yejin Choi. Is reinforcement learning (not) for natural language processing?: Benchmarks, baselines, and building blocks for natural language policy optimization. arXiv preprint arXiv:2210.01241, 2022. · arXiv
  24. John Schulman, Sergey Levine, Pieter Abbeel, Michael Jordan, and Philipp Moritz. Trust region policy optimization. In International conference on machine learning, pp. 1889-1897.
  25. PMLR, 2015a.
  26. John Schulman, Philipp Moritz, Sergey Levine, Michael Jordan, and Pieter Abbeel. High- dimensional continuous control using generalized advantage estimation. arXiv preprint arXiv:1506.02438, 2015b. · arXiv
  27. John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347, 2017. · arXiv
  28. Charlie Snell, Ilya Kostrikov, Yi Su, Mengjiao Yang, and Sergey Levine. Offline rl for natural language generation with implicit language q learning. arXiv preprint arXiv:2206.11871, 2022. · arXiv
  29. Feifan Song, Bowen Yu, Minghao Li, Haiyang Yu, Fei Huang, Yongbin Li, and Houfeng Wang. Preference ranking optimization for human alignment. arXiv preprint arXiv:2306.17492, 2023. · arXiv
  30. Richard S Sutton and Andrew G Barto. Reinforcement learning: An introduction. MIT press, 2018.
  31. Richard S Sutton, David McAllester, Satinder Singh, and Yishay Mansour. Policy gradient meth- ods for reinforcement learning with function approximation. Advances in neural information processing systems, 12, 1999.
  32. Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze Cheng, Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, et al. Lamda: Language models for dialog applications. arXiv preprint arXiv:2201.08239, 2022. · arXiv
  33. Michael Völske, Martin Potthast, Shahbaz Syed, and Benno Stein. Tl; dr: Mining reddit to learn automatic summarization. In Proceedings of the Workshop on New Frontiers in Summarization, pp. 59-63, 2017.
  34. Tianhao Wu, Yunchang Yang, Han Zhong, Liwei Wang, Simon Du, and Jiantao Jiao. Nearly optimal policy optimization with stable at any time guarantee. In International Conference on Machine Learning, pp. 24243-24265. PMLR, 2022.
  35. Jun Xu, Zeng Wei, Long Xia, Yanyan Lan, Dawei Yin, Xueqi Cheng, and Ji-Rong Wen. Reinforce- ment learning to rank with pairwise policy gradient. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, pp. 509-518, 2020.
  36. Yunchang Yang, Tianhao Wu, Han Zhong, Evrard Garcelon, Matteo Pirotta, Alessandro Lazaric, Liwei Wang, and Simon S Du. A reduction-based framework for conservative bandits and rein- forcement learning. arXiv preprint arXiv:2106.11692, 2021. · arXiv
  37. Zheng Yuan, Hongyi Yuan, Chuanqi Tan, Wei Wang, Songfang Huang, and Fei Huang. Rrhf: Rank responses to align language models with human feedback without tears. arXiv preprint arXiv:2304.05302, 2023. · arXiv
  38. Yisong Yue, Josef Broder, Robert Kleinberg, and Thorsten Joachims. The k-armed dueling bandits problem. Journal of Computer and System Sciences, 78(5):1538-1556, 2012.
  39. Rui Zheng, Shihan Dou, Songyang Gao, Wei Shen, Binghai Wang, Yan Liu, Senjie Jin, Qin Liu, Limao Xiong, Lu Chen, et al. Secrets of rlhf in large language models part i: Ppo. arXiv preprint arXiv:2307.04964, 2023. · arXiv
  40. Banghua Zhu, Jiantao Jiao, and Michael I Jordan. Principled reinforcement learning with human feedback from pairwise or k-wise comparisons. arXiv preprint arXiv:2301.11270, 2023a. · arXiv
  41. Banghua Zhu, Hiteshi Sharma, Felipe Vieira Frujeri, Shi Dong, Chenguang Zhu, Michael I Jordan, and Jiantao Jiao. Fine-tuning language models with advantage-induced policy alignment. arXiv preprint arXiv:2306.02231, 2023b. · arXiv
  42. Daniel M Ziegler, Nisan Stiennon, Jeffrey Wu, Tom B Brown, Alec Radford, Dario Amodei, Paul Christiano, and Geoffrey Irving. Fine-tuning language models from human preferences. arXiv preprint arXiv:1909.08593, 2019. · arXiv

Cited by

Sorted by citations

Metadata

Authors
Wu, T., Zhu, B., Zhang, R., Wen, Z., Ramchandran, K., & Jiao, J.
Affiliations
University of California, Berkeley (US)
ORCID
Venue
arXiv
Publisher
arXiv
Published
2023
ISSN
2331-8422 (ISSN-L)
DOI
10.48550/arxiv.2310.00212
arXiv
2310.00212
OpenAlex
W4387323351
Cited by
4
References
2
Citations by year
2024: 4
Type
posted content
Language
en
Open access
Licence
arxiv.org/licenses/nonexclusive-distrib/1.0/
Locations
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Reinforcement Learning in Robotics
  • Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
  • Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
Keywords
large language model alignment, reinforcement learning from human feedback

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review