Rafael Rafailov

EDBT 2026 Demo / reviewers in the wild / expert

dblp:272/5358 · DBLP profile ↗

← Back

18ranked-venue papers

3as first author

18since 2021 · last 2025

—ORCID · none

Domains — the database's venue-derived domains; a paper can count in several

Artificial intelligence and machine learning · 18 · 3 first-author · 18 since 2021Systems, architecture and hardware · 1 · 1 since 2021Graphics, computer vision, multimedia, augmented reality and games · 1 · 1 since 2021

Year	Publication	Venue	Position
2025	PERSONA: A Reproducible Testbed for Pluralistic Alignment abstract The rapid advancement of language models (LMs) necessitates robust alignment with diverse user values. However, current preference optimization approaches often fail to capture the plurality of user opinions, instead reinforcing majority viewpoints and marginalizing minority perspectives. We introduce PERSONA, a reproducible test bed designed to evaluate and improve pluralistic alignment of LMs. We procedurally generate diverse user profiles from US census data, resulting in 1,586 synthetic personas with varied demographic and idiosyncratic attributes. We then generate a large-scale evaluation dataset containing 3,868 prompts and 317,200 feedback pairs obtained from our synthetic personas. Leveraging this dataset, we systematically evaluate LM capabilities in role-playing diverse users, verified through human judges, and the establishment of both a benchmark, PERSONA Bench, for pluralistic alignment approaches as well as an extensive dataset to create new and future benchmarks. Louis Castricato, Nathan Lile, Rafael Rafailov, Jan-Philipp Fränken, Chelsea Finn	COLING	3
2025	Collapse or Thrive: Perils and Promises of Synthetic Data in a Self-Generating World abstract What happens when generative machine learning models are pretrained on web-scale datasets containing data generated by earlier models? Some prior work warns of “model collapse” as the web is overwhelmed by synthetic data; other work suggests the problem can be contained (i.e. collapse can be avoided) by managing how available data are used in pretraining. In this paper, we report experiments on three ways of using data (training-workflows), across three generative model task-settings (multivariate Gaussian estimation, kernel density estimation, and language-model fine-tuning) to further confirm the possibility of containment: (a) we confirm that the training-workflow of {\it replacing} all real data by successive generations of purely synthetic data indeed suffers model collapse in all task-settings studied; (b) we consider the training-workflow of {\it accumulating} synthetic data alongside real data and training on all data combined and confirming that, although the proportion of real data eventually becomes zero, models remain stable and their test losses do not diverge under this training-workflow; (c) we consider a training-workflow where real and synthetic data accumulate together but successive generations of pretraining are constrained to use fixed-size data subsets each generation. In this workflow, we observe slow and gradual rather than explosive degradation of test loss performance across generations. Our insights are particularly important when forecasting whether future frontier generative models will collapse or thrive, and our results open avenues for empirically and mathematically studying the context-dependent value of synthetic data. Joshua Kazdan, Rylan Schaeffer, Apratim Dey, Matthias Gerstgrasser, Rafael Rafailov, David L. Donoho, Oluwasanmi Koyejo	ICML	5
2025	MJ-Bench: Is Your Multimodal Reward Model Really a Good Judge for Text-to-Image Generation? abstract While text-to-image models like GPT-4o-Image and FLUX are rapidly proliferating, they often encounter challenges such as hallucination, bias, and the production of unsafe, low-quality output. To effectively address these issues, it is crucial to align these models with desired behaviors based on feedback from a multimodal judge. Despite their significance, current multimodal judges frequently undergo inadequate evaluation of their capabilities and limitations, potentially leading to misalignment and unsafe fine-tuning outcomes. To address this issue, we introduce MJ-Bench, a novel benchmark which incorporates a comprehensive preference dataset to evaluate multimodal judges in providing feedback for image generation models across six key perspectives: alignment, safety, image quality, bias, composition, and visualization. Specifically, we evaluate a large variety of multimodal judges including smaller-sized CLIP-based scoring models, open-source VLMs, and close-source VLMs on each decomposed subcategory of our preference dataset. Experiments reveal that close-source VLMs generally provide better feedback, with GPT-4o outperforming other judges in average. Compared with open-source VLMs, smaller-sized scoring models can provide better feedback regarding text-image alignment and image quality, while VLMs provide more accurate feedback regarding safety and generation bias due to their stronger reasoning capabilities. Further studies in feedback scale reveal that VLM judges can generally provide more accurate and stable feedback in natural language than numerical scales. Notably, human evaluations on end-to-end and fine-tuned models using separate feedback from these multimodal judges provide similar conclusions, further confirming the effectiveness of MJ-Bench. Zhaorun Chen, Zichen Wen, Yichao Du, Yiyang Zhou, Chenhang Cui, Siwei Han, Jen Weng, Chaoqi Wang, Zhengwei Tong, Leria Huang, Canyu Chen, Haoqin Tu, Qinghao Ye, Zhihong Zhu 0001, Zhuokai Zhao, Rafael Rafailov, Chelsea Finn, Huaxiu Yao	NeurIPS	18
2025	MJ-Video: Benchmarking and Rewarding Video Generation with Fine-Grained Video Preference abstract Recent advancements in video generation have significantly improved the ability to synthesize videos from text instructions. However, existing models still struggle with key challenges such as instruction misalignment, content hallucination, safety concerns, and generation bias. To address these limitations, we introduce MJ-BENCH-VIDEO, a large-scale video preference benchmark designed to evaluate video generation across five critical aspects: Alignment, Safety, Fineness, Coherence & Consistency, and Bias & Fairness. This benchmark further incorporates 28 fine-grained criteria to provide a comprehensive evaluation of video preference. Building upon this dataset, we propose MJ-VIDEO, a Mixture-of-Experts (MoE)-based video reward model designed to deliver fine-grained reward. MJ-VIDEO can dynamically select relevant experts to accurately judge the preference based on the input text-video pair. This architecture enables more precise and adaptable preference judgments. Through extensive benchmarking on MJ-BENCH-VIDEO, we analyze the limitations of existing video reward models and demonstrate the superior performance of MJ-VIDEO in video preference assessment, achieving 17.58% and 15.87% improvements in overall and fine-grained preference judgments, respectively. Additionally, MJ-VIDEO is able to improve the alignment performance in video generation via preference fine-tuning. Haibo Tong, Zhaoyang Wang 0004, Zhaorun Chen, Haonian Ji, Shi Qiu 0016, Siwei Han, Kexin Geng, Zhongkai Xue, Yiyang Zhou, Peng Xia 0005, Mingyu Ding, Rafael Rafailov, Chelsea Finn, Huaxiu Yao	NeurIPS	12
2024	Diffusion Model Alignment Using Direct Preference Optimization abstract Large language models (LLMs) are fine-tuned using human comparison data with Reinforcement Learning from Human Feedback (RLHF) methods to make them better aligned with users' preferences. In contrast to LLMs, human preference learning has not been widely explored in text-to-image diffusion models; the best existing approach is to fine-tune a pretrained model using carefully curated high quality images and captions to improve visual appeal and text alignment. We propose Diffusion-DPO, a method to align diffusion models to human preferences by directly optimizing on human comparison data. Diffusion-DPO is adapted from the recently developed Direct Preference Optimization (DPO) [36], a simpler alternative to RLHF which directly optimizes a policy that best satisfies human preferences under a classification objective. We re-formulate DPO to account for a diffusion model notion of likelihood, utilizing the evidence lower bound to derive a differentiable objective. Using the Pick-a-Pic dataset of 851K crowdsourced pairwise preferences, we fine-tune the base model of the state-of-the-art Stable Diffusion XL (SDXL)-1.0 model with Diffusion-DPO. Our fine-tuned base model significantly outperforms both base SDXL-1.0 and the larger SDXL-1.0 model consisting of an additional refinement model in human evaluation, improving visual appeal and prompt alignment. We also develop a variant that uses AI feedback and has comparable performance to training on human preferences, opening the door for scaling of diffusion model alignment methods. Bram Wallace, Meihua Dang, Rafael Rafailov, Linqi Zhou, Aaron Lou, Senthil Purushwalkam, Stefano Ermon, Caiming Xiong, Shafiq R. Joty	CVPR	3
2024	Contrastive Preference Learning: Learning from Human Feedback without Reinforcement Learning abstract Reinforcement Learning from Human Feedback (RLHF) has emerged as a popular paradigm for aligning models with human intent. Typically RLHF algorithms operate in two phases: first, use human preferences to learn a reward function and second, align the model by optimizing the learned reward via reinforcement learning (RL). This paradigm assumes that human preferences are distributed according to reward, but recent work suggests that they instead follow the \emph{regret} under the user's optimal policy. Thus, learning a reward function from feedback is not only based on a flawed assumption of human preference, but also leads to unwieldy optimization challenges that stem from policy gradients or bootstrapping in the RL phase. Because of these optimization challenges, contemporary RLHF methods restrict themselves to contextual bandit settings (e.g., as in large language models) or limit observation dimensionality (e.g., state-based robotics). We overcome these limitations by introducing a new family of algorithms for optimizing behavior from human feedback using the \textit{regret}-based model of human preferences. Using the principle of maximum entropy, we derive \fullname (\abv), an algorithm for learning optimal policies from preferences without learning reward functions, circumventing the need for RL. \abv is fully off-policy, uses only a simple contrastive objective, and can be applied to arbitrary MDPs. This enables \abv to elegantly scale to high-dimensional and sequential RLHF problems while being simpler than prior methods. Joey Hejna, Rafael Rafailov, Harshit Sikchi, Chelsea Finn, Scott Niekum, W. Bradley Knox, Dorsa Sadigh	ICLR	2
2024	An Emulator for Fine-tuning Large Language Models using Small Language Models abstract Widely used language models (LMs) are typically built by scaling up a two-stage training pipeline: a pre-training stage that uses a very large, diverse dataset of text and a fine-tuning (sometimes, 'alignment') stage that uses targeted examples or other specifications of desired behaviors. While it has been hypothesized that knowledge and skills come from pre-training, and fine-tuning mostly filters this knowledge and skillset, this intuition has not been extensively tested. To aid in doing so, we introduce a novel technique for decoupling the knowledge and skills gained in these two stages, enabling a direct answer to the question, What would happen if we combined the knowledge learned by a large model during pre-training with the knowledge learned by a small model during fine-tuning (or vice versa)? Using an RL-based framework derived from recent developments in learning from human preferences, we introduce emulated fine-tuning (EFT), a principled and practical method for sampling from a distribution that approximates (or 'emulates') the result of pre-training and fine-tuning at different scales. Our experiments with EFT show that scaling up fine-tuning tends to improve helpfulness, while scaling up pre-training tends to improve factuality. Beyond decoupling scale, we show that EFT enables test-time adjustment of competing behavioral traits like helpfulness and harmlessness without additional training. Finally, a special case of emulated fine-tuning, which we call LM up-scaling, avoids resource-intensive fine-tuning of large pre-trained models by ensembling them with small fine-tuned models, essentially emulating the result of fine-tuning the large pre-trained model. Up-scaling consistently improves helpfulness and factuality of instruction-following models in the Llama, Llama-2, and Falcon families, without additional hyperparameters or training. For reference implementation, see [https://github.com/eric-mitchell/emulated-fine-tuning](https://github.com/eric-mitchell/emulated-fine-tuning). Eric Mitchell, Rafael Rafailov, Archit Sharma, Chelsea Finn, Christopher D. Manning	ICLR	2
2024	Language Model Detectors Are Easily Optimized Against abstract The fluency and general applicability of large language models (LLMs) has motivated significant interest in detecting whether a piece of text was written by a language model. While both academic and commercial detectors have been deployed in some settings, particularly education, other research has highlighted the fragility of these systems. In this paper, we demonstrate a data-efficient attack that fine-tunes language models to confuse existing detectors, leveraging recent developments in reinforcement learning of language models. We use the `human-ness' score (often just a log probability) of various open-source and commercial detectors as a reward function for reinforcement learning, subject to a KL-divergence constraint that the resulting model does not differ significantly from the original. For a 7B parameter Llama-2 model, fine-tuning for under a day reduces the AUROC of the OpenAI RoBERTa-Large detector from 0.84 to 0.63, while perplexity on OpenWebText increases from 8.7 to only 9.0; with a larger perplexity budget, we can drive AUROC to 0.30 (worse than random). Similar to traditional adversarial attacks, we find that this increase in 'detector evasion' generalizes to other detectors not used during training. In light of our empirical results, we advise against continued reliance on LLM-generated text detectors. Models, datasets, and selected experiment code will be released at https://github.com/charlottttee/llm-detector-evasion. Charlotte Nicks, Eric Mitchell, Rafael Rafailov, Archit Sharma, Christopher D. Manning, Chelsea Finn, Stefano Ermon	ICLR	3
2024	Preference Fine-Tuning of LLMs Should Leverage Suboptimal, On-Policy Data abstract Learning from preference labels plays a crucial role in fine-tuning large language models — this is done via supervised learning, on-policy reinforcement learning (RL), or contrastive learning. Different methods come with different implementation tradeoffs, and existing empirical findings present different conclusions, for instance, some results show that online RL is quite important to attain good fine-tuning results, while others find offline methods sufficient. This raises a question: what kind of approaches are important for fine-tuning with preference data and why? In this paper, we answer this question by performing a rigorous analysis of a number of fine-tuning techniques on didactic and full-scale LLM problems. Our main finding is that approaches that use on-policy sampling and attempt to push down the likelihood on certain responses (i.e., employ a ”negative gradient”) outperform offline and maximum likelihood objectives. We conceptualize our insights and unify methods that use on-policy sampling or negative gradient under a notion of mode-seeking objectives for categorical distributions. Mode-seeking objectives are able to alter probability mass on specific bins of a categorical distribution at a fast rate compared to maximum likelihood, allowing them to relocate masses across bins more effectively. Our analysis prescribes actionable insights for preference fine-tuning of LLMs and informs how data should be collected for maximal improvement. Fahim Tajwar, Anikait Singh, Archit Sharma, Rafael Rafailov, Jeff G. Schneider, Tengyang Xie, Stefano Ermon, Chelsea Finn, Aviral Kumar	ICML	4
2024	Open X-Embodiment: Robotic Learning Datasets and RT-X Models : Open X-Embodiment Collaboration abstract Large, high-capacity models trained on diverse datasets have shown remarkable successes on efficiently tackling downstream applications. In domains from NLP to Computer Vision, this has led to a consolidation of pretrained models, with general pretrained backbones serving as a starting point for many applications. Can such a consolidation happen in robotics? Conventionally, robotic learning methods train a separate model for every application, every robot, and even every environment. Can we instead train "generalist" X-robot policy that can be adapted efficiently to new robots, tasks, and environments? In this paper, we provide datasets in standardized data formats and models to make it possible to explore this possibility in the context of robotic manipulation, alongside experimental results that provide an example of effective X-robot policies. We assemble a dataset from 22 different robots collected through a collaboration between 21 institutions, demonstrating 527 skills (160266 tasks). We show that a high-capacity model trained on this data, which we call RT-X, exhibits positive transfer and improves the capabilities of multiple robots by leveraging experience from other platforms. The project website is robotics-transformer-x.github.io. Abigail O'Neill, Abhiram Maddukuri, Abhishek Gupta 0004, Abhishek Padalkar, Abraham Lee, Acorn Pooley, Agrim Gupta, Ajay Mandlekar, Ajinkya Jain, Albert Tung, Alex Bewley, Alex Irpan, Alexander Khazatsky, Anant Rai, Anchit Gupta, Andrew E. Wang, Anikait Singh, Animesh Garg, Aniruddha Kembhavi, Annie Xie, Anthony Brohan, Antonin Raffin, Archit Sharma, Arefeh Yavary, Arhan Jain, Ashwin Balakrishna, Ayzaan Wahid, Ben Burgess-Limerick, Bernhard Schölkopf, Blake Wulfe, Brian Ichter, Cewu Lu, Charles Xu 0003, Charlotte Le, Chelsea Finn, Chen Wang 0053, Chenfeng Xu, Cheng Chi 0001, Chenguang Huang, Christine Chan, Christopher Agia, Chuer Pan, Chuyuan Fu, Coline Devin, Danfei Xu, Daniel Morton, Danny Drieß, Daphne Chen, Deepak Pathak, Dhruv Shah, Dieter Büchler, Dinesh Jayaraman, Dmitry Kalashnikov, Dorsa Sadigh, Edward Johns, Ethan Paul Foster, Fangchen Liu, Federico Ceola, Fei Xia 0002, Feiyu Zhao, Freek Stulp, Gaoyue Zhou, Gaurav S. Sukhatme, Gautam Salhotra, Gilbert Feng, Giulio Schiavi, Glen Berseth, Gregory Kahn, Guanzhi Wang, Hao Su 0001, Haoshu Fang, Henghui Bao, Heni Ben Amor, Henrik I. Christensen, Hiroki Furuta, Homer Walke, Hongjie Fang, Huy Ha, Igor Mordatch, Ilija Radosavovic, Isabel Leal, Jacky Liang, Jad Abou-Chakra, Jaehyung Kim 0001, Jaimyn Drake, Jan Peters 0001, Jan Schneider 0007, Jasmine Hsu, Jeannette Bohg, Jeffrey T. Bingham, Jensen Gao, Jiaheng Hu, Jiajun Wu 0001, Jiankai Sun, Jianlan Luo, Jiayuan Gu, Jie Tan 0001, Jihoon Oh, Jimmy Wu, Jingpei Lu, Jitendra Malik, João Silvério, Joey Hejna, Jonathan Booher, Jonathan Tompson, Jonathan Yang, Jordi Salvador, Joseph J. Lim, Junhyek Han, Kanishka Rao, Karl Pertsch, Karol Hausman, Keegan Go, Keerthana Gopalakrishnan, Kenneth Y. Goldberg, Kendra Byrne, Kenneth Oslund, Kento Kawaharazuka, Kevin Black, Kevin Zhang 0002, Kiana Ehsani, Kiran Lekkala, Kirsty Ellis, Krishan Rana, Krishnan Srinivasan, Kuan Fang, Kunal Pratap Singh, Kuo-Hao Zeng, Kyle Hatch, Kyle Hsu, Laurent Itti, Yunliang Chen 0001, Lerrel Pinto, Li Fei-Fei 0001, Liam Tan, Linxi Fan, Lionel Ott, Lisa Lee, Luca Weihs, Magnum Chen, Marion Lepert, Marius Memmel, Masayoshi Tomizuka, Masha Itkina, Mateo Guaman Castro, Max Spero, Maximilian Du, Michael Ahn, Michael C. Yip, Mingtong Zhang 0003, Mingyu Ding, Minho Heo, Mohan Kumar Srirama, Mohit Sharma 0001, Moo Jin Kim, Naoaki Kanazawa, Nicklas Hansen 0001, Nicolas Heess, Nikhil J. Joshi, Niko Sünderhauf, Norman Di Palo, Nur Muhammad Shafiullah, Oier Mees, Oliver Kroemer, Osbert Bastani, Pannag R. Sanketi, Patrick Tree Miller, Patrick Yin, Paul Wohlhart, Peng Xu 0010, Peter David Fagan, Peter Mitrano, Pierre Sermanet, Pieter Abbeel, Priya Sundaresan, Qiuyu Chen, Rafael Rafailov, Ria Doshi, Roberto Martin Martin, Rohan Baijal, Rosario Scalise, Rose Hendrix, Roy Lin, Runjia Qian, Russell Mendonca, Rutav Shah, Ryan Hoque, Ryan Julian, Samuel Bustamante-Gomez, Sean Kirmani, Sergey Levine, Sherry Moore, Shikhar Bahl, Shivin Dass, Shubham D. Sonawani, Shuran Song, Sichun Xu, Siddhant Haldar, Siddharth Karamcheti, Simeon Adebola, Simon Guist, Soroush Nasiriany, Stefan Schaal, Stefan Welker, Stephen Tian, Subramanian Ramamoorthy, Sudeep Dasari, Suneel Belkhale, Sungjae Park, Suraj Nair 0003, Suvir Mirchandani, Takayuki Osa, Tanmay Gupta, Tatsuya Harada, Tatsuya Matsushima, Ted Xiao, Thomas Kollar, Tianhe Yu, Tianli Ding, Todor Davchev, Tony Z. Zhao, Travis Armstrong, Trevor Darrell, Trinity Chung, Vidhi Jain, Vincent Vanhoucke, Wolfram Burgard, Xiaolong Wang 0004, Xinghao Zhu, Xinyang Geng, Liangwei Xu, Yecheng Jason Ma 0001, Yejin Kim 0003, Yevgen Chebotar, Yilin Wu 0003, Yonatan Bisk, Yoonyoung Cho, Youngwoon Lee, Yuchen Cui, Yueh-Hua Wu, Yujin Tang, Yuke Zhu, Yunchu Zhang, Yunfan Jiang 0001, Yunshuang Li, Yunzhu Li, Yusuke Iwasawa, Yutaka Matsuo, Zehan Ma, Zichen Jeff Cui, Zichen Zhang 0016, Zipeng Lin	ICRA	189
2024	Self-Supervised Alignment with Mutual Information: Learning to Follow Principles without Preference Labels abstract When prompting a language model (LM), users often expect the model to adhere to a set of behavioral principles across diverse tasks, such as producing insightful content while avoiding harmful or biased language. Instilling such principles (i.e., a constitution) into a model is resource-intensive, technically challenging, and generally requires human preference labels or examples. We introduce SAMI, an iterative algorithm that finetunes a pretrained language model (without requiring preference labels or demonstrations) to increase the conditional mutual information between constitutions and self-generated responses given queries from a dataset. On single-turn dialogue and summarization, a SAMI-trained mistral-7b outperforms the initial pretrained model, with win rates between 66% and 77%. Strikingly, it also surpasses an instruction-finetuned baseline (mistral-7b-instruct) with win rates between 55% and 57% on single-turn dialogue. SAMI requires a model that writes the principles. To avoid dependence on strong models for writing principles, we align a strong pretrained model (mixtral-8x7b) using constitutions written by a weak instruction-finetuned model (mistral-7b-instruct), achieving a 65% win rate on summarization. Finally, we investigate whether SAMI generalizes to diverse summarization principles (e.g., "summaries should be scientific") and scales to stronger models (llama3-70b), finding that it achieves win rates of up to 68% for learned and 67% for held-out principles compared to the base model. Our results show that a pretrained LM can learn to follow constitutions without using preference labels, demonstrations, or human oversight. Jan-Philipp Fränken, Eric Zelikman, Rafael Rafailov, Kanishk Gandhi, Tobias Gerstenberg, Noah D. Goodman	NeurIPS	3
2024	Scaling Laws for Reward Model Overoptimization in Direct Alignment Algorithms abstract Reinforcement Learning from Human Feedback (RLHF)has been crucial to the recent success of Large Language Models (LLMs), however it is often a complex and brittle process. In the classical RLHF framework, a reward model is first trained to represent human preferences, which is in turn used by an online reinforcement learning (RL) algorithm to optimized the LLM. A prominent issue with such methods is reward over-optimization or reward hacking, where the performance as measured by the learned proxy reward model increases, but the true model quality plateaus or even deteriorates. Direct Alignment Algorithms (DDAs), such as Direct Preference Optimization (DPO) have emerged as alternatives to the classical RLHF pipeline. However, despite not training a separate proxy reward model or using RL, they still commonly deteriorate from over-optimization. While the so-called reward hacking phenomenon is not well-defined for DAAs, we still uncover similar trends: at higher KL-budgets, DAA algorithms exhibit similar degradation patters to their classic RLHF counterparts. In particular, we find that DAA methods deteriorate not only across a wide range of KL-budgets, but also often before even a single epoch of the dataset is completed. Through extensive empirical experimentation this work formulates the reward over-optimization or hacking problem for DAAs and explores its consequences across objectives, training regimes, and model scales. Rafael Rafailov, Yaswanth Chittepu, Ryan Park, Harshit Sikchi, Joey Hejna, W. Bradley Knox, Chelsea Finn, Scott Niekum	NeurIPS	1
2023	Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedback abstract Katherine Tian, Eric Mitchell, Allan Zhou, Archit Sharma, Rafael Rafailov, Huaxiu Yao, Chelsea Finn, Christopher Manning. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023. Katherine Tian, Eric Mitchell, Allan Zhou, Archit Sharma, Rafael Rafailov, Huaxiu Yao, Chelsea Finn, Christopher D. Manning	EMNLP	5
2023	Direct Preference Optimization: Your Language Model is Secretly a Reward Model abstract While large-scale unsupervised language models (LMs) learn broad world knowledge and some reasoning skills, achieving precise control of their behavior is difficult due to the completely unsupervised nature of their training. Existing methods for gaining such steerability collect human labels of the relative quality of model generations and fine-tune the unsupervised LM to align with these preferences, often with reinforcement learning from human feedback (RLHF). However, RLHF is a complex and often unstable procedure, first fitting a reward model that reflects the human preferences, and then fine-tuning the large unsupervised LM using reinforcement learning to maximize this estimated reward without drifting too far from the original model. In this paper, we leverage a mapping between reward functions and optimal policies to show that this constrained reward maximization problem can be optimized exactly with a single stage of policy training, essentially solving a classification problem on the human preference data. The resulting algorithm, which we call Direct Preference Optimization (DPO), is stable, performant, and computationally lightweight, eliminating the need for fitting a reward model, sampling from the LM during fine-tuning, or performing significant hyperparameter tuning. Our experiments show that DPO can fine-tune LMs to align with human preferences as well as or better than existing methods. Notably, fine-tuning with DPO exceeds RLHF's ability to control sentiment of generations and improves response quality in summarization and single-turn dialogue while being substantially simpler to implement and train. Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D. Manning, Stefano Ermon, Chelsea Finn	NeurIPS	1
2022	Vision-Based Manipulators Need to Also See from Their Hands Kyle Hsu, Moo Jin Kim, Rafael Rafailov, Jiajun Wu 0001, Chelsea Finn	ICLR	3
2021	Offline Meta-Reinforcement Learning with Advantage Weighting abstract This paper introduces the offline meta-reinforcement learning (offline meta-RL) problem setting and proposes an algorithm that performs well in this setting. Offline meta-RL is analogous to the widely successful supervised learning strategy of pre-training a model on a large batch of fixed, pre-collected data (possibly from various tasks) and fine-tuning the model to a new task with relatively little data. That is, in offline meta-RL, we meta-train on fixed, pre-collected data from several tasks and adapt to a new task with a very small amount (less than 5 trajectories) of data from the new task. By nature of being offline, algorithms for offline meta-RL can utilize the largest possible pool of training data available and eliminate potentially unsafe or costly data collection during meta-training. This setting inherits the challenges of offline RL, but it differs significantly because offline RL does not generally consider a) transfer to new tasks or b) limited data from the test task, both of which we face in offline meta-RL. Targeting the offline meta-RL setting, we propose Meta-Actor Critic with Advantage Weighting (MACAW). MACAW is an optimization-based meta-learning algorithm that uses simple, supervised regression objectives for both the inner and outer loop of meta-training. On offline variants of common meta-RL benchmarks, we empirically find that this approach enables fully offline meta-reinforcement learning and achieves notable gains over prior methods. Eric Mitchell, Rafael Rafailov, Xue Bin Peng, Sergey Levine, Chelsea Finn	ICML	2
2021	Visual Adversarial Imitation Learning using Variational Models abstract Reward function specification, which requires considerable human effort and iteration, remains a major impediment for learning behaviors through deep reinforcement learning. In contrast, providing visual demonstrations of desired behaviors presents an easier and more natural way to teach agents. We consider a setting where an agent is provided a fixed dataset of visual demonstrations illustrating how to perform a task, and must learn to solve the task using the provided demonstrations and unsupervised environment interactions. This setting presents a number of challenges including representation learning for visual observations, sample complexity due to high dimensional spaces, and learning instability due to the lack of a fixed reward or learning signal. Towards addressing these challenges, we develop a variational model-based adversarial imitation learning (V-MAIL) algorithm. The model-based approach provides a strong signal for representation learning, enables sample efficiency, and improves the stability of adversarial training by enabling on-policy learning. Through experiments involving several vision-based locomotion and manipulation tasks, we find that V-MAIL learns successful visuomotor policies in a sample-efficient manner, has better stability compared to prior work, and also achieves higher asymptotic performance. We further find that by transferring the learned models, V-MAIL can learn new tasks from visual demonstrations without any additional environment interactions. All results including videos can be found online at https://sites.google.com/view/variational-mail Rafael Rafailov, Tianhe Yu, Aravind Rajeswaran, Chelsea Finn	NeurIPS	1
2021	COMBO: Conservative Offline Model-Based Policy Optimization abstract Model-based reinforcement learning (RL) algorithms, which learn a dynamics model from logged experience and perform conservative planning under the learned model, have emerged as a promising paradigm for offline reinforcement learning (offline RL). However, practical variants of such model-based algorithms rely on explicit uncertainty quantification for incorporating conservatism. Uncertainty estimation with complex models, such as deep neural networks, can be difficult and unreliable. We empirically find that uncertainty estimation is not accurate and leads to poor performance in certain scenarios in offline model-based RL. We overcome this limitation by developing a new model-based offline RL algorithm, COMBO, that trains a value function using both the offline dataset and data generated using rollouts under the model while also additionally regularizing the value function on out-of-support state-action tuples generated via model rollouts. This results in a conservative estimate of the value function for out-of-support state-action tuples, without requiring explicit uncertainty estimation. Theoretically, we show that COMBO satisfies a policy improvement guarantee in the offline setting. Through extensive experiments, we find that COMBO attains greater performance compared to prior offline RL on problems that demand generalization to related but previously unseen tasks, and also consistently matches or outperforms prior offline RL methods on widely studied offline RL benchmarks, including image-based tasks. Tianhe Yu, Aviral Kumar, Rafael Rafailov, Aravind Rajeswaran, Sergey Levine, Chelsea Finn	NeurIPS	3