Jonathan Chang

Jonathan D. Chang, Kiante Brantley, Rajkumar Ramamurthy et al.

Reinforcement learning (RL) has emerged as a powerful paradigm for fine-tuning Large Language Models (LLMs) for text generation. In particular, recent LLMs such as ChatGPT and GPT-4 can engage in fluent conversations with users after finetuning with RL. Capitalizing on key properties of text generation, we seek to investigate RL algorithms beyond general purpose algorithms like Proximal Policy Optimization (PPO). In particular, we extend RL algorithms to allow them to interact with a dynamic black-box guide LLM and propose RL with guided feedback (RLGF), a suite of RL algorithms for LLM fine-tuning. We provide two ways for the guide LLM to interact with the LLM to be optimized for maximizing rewards. The guide LLM can generate text which serves as additional starting states for the RL optimization procedure. The guide LLM can also be used to complete the partial sentences generated by the LLM that is being optimized, treating the guide LLM as an expert to imitate and surpass eventually. We experiment on the IMDB positive sentiment, CommonGen, and TL;DR summarization tasks. We show that our RL algorithms achieve higher performance than supervised learning (SL) and the RL baseline PPO, demonstrating the benefit of interaction with the guide LLM. On both CommonGen and TL;DR, we not only outperform our SL baselines but also improve upon PPO across a variety of metrics beyond the one we optimized for. Our code can be found at https://github.com/Cornell-RL/tril.

Jonathan D. Chang, Kaiwen Wang, Nathan Kallus et al.

We study representation learning for Offline Reinforcement Learning (RL), focusing on the important task of Offline Policy Evaluation (OPE). Recent work shows that, in contrast to supervised learning, realizability of the Q-function is not enough for learning it. Two sufficient conditions for sample-efficient OPE are Bellman completeness and coverage. Prior work often assumes that representations satisfying these conditions are given, with results being mostly theoretical in nature. In this work, we propose BCRL, which directly learns from data an approximately linear Bellman complete representation with good coverage. With this learned representation, we perform OPE using Least Square Policy Evaluation (LSPE) with linear functions in our learned representation. We present an end-to-end theoretical analysis, showing that our two-stage algorithm enjoys polynomial sample complexity provided some representation in the rich class considered is linear Bellman complete. Empirically, we extensively evaluate our algorithm on challenging, image-based continuous control tasks from the Deepmind Control Suite. We show our representation enables better OPE compared to previous representation learning methods developed for off-policy RL (e.g., CURL, SPR). BCRL achieve competitive OPE error with the state-of-the-art method Fitted Q-Evaluation (FQE), and beats FQE when evaluating beyond the initial state distribution. Our ablations show that both linear Bellman complete and coverage components of our method are crucial.

Ge Gao, Jonathan D. Chang, Claire Cardie et al.

Text retrieval plays a crucial role in incorporating factual knowledge for decision making into language processing pipelines, ranging from chat-based web search to question answering systems. Current state-of-the-art text retrieval models leverage pre-trained large language models (LLMs) to achieve competitive performance, but training LLM-based retrievers via typical contrastive losses requires intricate heuristics, including selecting hard negatives and using additional supervision as learning signals. This reliance on heuristics stems from the fact that the contrastive loss itself is heuristic and does not directly optimize the downstream metrics of decision quality at the end of the processing pipeline. To address this issue, we introduce Neural PG-RANK, a novel training algorithm that learns to rank by instantiating a LLM as a Plackett-Luce ranking policy. Neural PG-RANK provides a principled method for end-to-end training of retrieval models as part of larger decision systems via policy gradient, with little reliance on complex heuristics, and it effectively unifies the training objective with downstream decision-making quality. We conduct extensive experiments on various text retrieval benchmarks. The results demonstrate that when the training objective aligns with the evaluation setup, Neural PG-RANK yields remarkable in-domain performance improvement, with substantial out-of-domain generalization to some critical datasets employed in downstream question answering tasks.

Jonathan D. Chang, Wenhao Zhan, Owen Oertell et al.

Reinforcement Learning (RL) from Human Preference-based feedback is a popular paradigm for fine-tuning generative models, which has produced impressive models such as GPT-4 and Claude3 Opus. This framework often consists of two steps: learning a reward model from an offline preference dataset followed by running online RL to optimize the learned reward model. In this work, leveraging the idea of reset, we propose a new RLHF algorithm with provable guarantees. Motivated by the fact that offline preference dataset provides informative states (i.e., data that is preferred by the labelers), our new algorithm, Dataset Reset Policy Optimization (DR-PO), integrates the existing offline preference dataset into the online policy training procedure via dataset reset: it directly resets the policy optimizer to the states in the offline dataset, instead of always starting from the initial state distribution. In theory, we show that DR-PO learns to perform at least as good as any policy that is covered by the offline dataset under general function approximation with finite sample complexity. In experiments, we demonstrate that on both the TL;DR summarization and the Anthropic Helpful Harmful (HH) dataset, the generation from DR-PO is better than that from Proximal Policy Optimization (PPO) and Direction Preference Optimization (DPO), under the metric of GPT4 win-rate. Code for this work can be found at https://github.com/Cornell-RL/drpo.

Jin Peng Zhou, Kaiwen Wang, Jonathan Chang et al.

Reinforcement learning (RL) post-training is crucial for LLM alignment and reasoning, but existing policy-based methods, such as PPO and DPO, can fall short of fixing shortcuts inherited from pre-training. In this work, we introduce $Q\sharp$, a value-based algorithm for KL-regularized RL that guides the reference policy using the optimal regularized $Q$ function. We propose to learn the optimal $Q$ function using distributional RL on an aggregated online dataset. Unlike prior value-based baselines that guide the model using unregularized $Q$-values, our method is theoretically principled and provably learns the optimal policy for the KL-regularized RL problem. Empirically, $Q\sharp$ outperforms prior baselines in math reasoning benchmarks while maintaining a smaller KL divergence to the reference policy. Theoretically, we establish a reduction from KL-regularized RL to no-regret online learning, providing the first bounds for deterministic MDPs under only realizability. Thanks to distributional RL, our bounds are also variance-dependent and converge faster when the reference policy has small variance. In sum, our results highlight $Q\sharp$ as an effective approach for post-training LLMs, offering both improved performance and theoretical guarantees. The code can be found at https://github.com/jinpz/q_sharp.

Owen Oertell, Jonathan D. Chang, Yiyi Zhang et al.

Reinforcement learning (RL) has improved guided image generation with diffusion models by directly optimizing rewards that capture image quality, aesthetics, and instruction following capabilities. However, the resulting generative policies inherit the same iterative sampling process of diffusion models that causes slow generation. To overcome this limitation, consistency models proposed learning a new class of generative models that directly map noise to data, resulting in a model that can generate an image in as few as one sampling iteration. In this work, to optimize text-to-image generative models for task specific rewards and enable fast training and inference, we propose a framework for fine-tuning consistency models via RL. Our framework, called Reinforcement Learning for Consistency Model (RLCM), frames the iterative inference process of a consistency model as an RL procedure. Comparing to RL finetuned diffusion models, RLCM trains significantly faster, improves the quality of the generation measured under the reward objectives, and speeds up the inference procedure by generating high quality images with as few as two inference steps. Experimentally, we show that RLCM can adapt text-to-image consistency models to objectives that are challenging to express with prompting, such as image compressibility, and those derived from human feedback, such as aesthetic quality. Our code is available at https://rlcm.owenoertell.com.

Jonathan D. Chang, Masatoshi Uehara, Dhruv Sreenivas et al.

This paper studies offline Imitation Learning (IL) where an agent learns to imitate an expert demonstrator without additional online environment interactions. Instead, the learner is presented with a static offline dataset of state-action-next state transition triples from a potentially less proficient behavior policy. We introduce Model-based IL from Offline data (MILO): an algorithmic framework that utilizes the static dataset to solve the offline IL problem efficiently both in theory and in practice. In theory, even if the behavior policy is highly sub-optimal compared to the expert, we show that as long as the data from the behavior policy provides sufficient coverage on the expert state-action traces (and with no necessity for a global coverage over the entire state-action space), MILO can provably combat the covariate shift issue in IL. Complementing our theory results, we also demonstrate that a practical implementation of our approach mitigates covariate shift on benchmark MuJoCo continuous control tasks. We demonstrate that with behavior policies whose performances are less than half of that of the expert, MILO still successfully imitates with an extremely low number of expert state-action pairs while traditional offline IL method such as behavior cloning (BC) fails completely. Source code is provided at https://github.com/jdchang1/milo.

Rahul Kidambi, Jonathan Chang, Wen Sun

This paper studies Imitation Learning from Observations alone (ILFO) where the learner is presented with expert demonstrations that consist only of states visited by an expert (without access to actions taken by the expert). We present a provably efficient model-based framework MobILE to solve the ILFO problem. MobILE involves carefully trading off strategic exploration against imitation - this is achieved by integrating the idea of optimism in the face of uncertainty into the distribution matching imitation learning (IL) framework. We provide a unified analysis for MobILE, and demonstrate that MobILE enjoys strong performance guarantees for classes of MDP dynamics that satisfy certain well studied notions of structural complexity. We also show that the ILFO problem is strictly harder than the standard IL problem by presenting an exponential sample complexity separation between IL and ILFO. We complement these theoretical results with experimental simulations on benchmark OpenAI Gym tasks that indicate the efficacy of MobILE. Code for implementing the MobILE framework is available at https://github.com/rahulkidambi/MobILE-NeurIPS2021.

Zhaolin Gao, Jonathan D. Chang, Wenhao Zhan et al.

While originally developed for continuous control problems, Proximal Policy Optimization (PPO) has emerged as the work-horse of a variety of reinforcement learning (RL) applications, including the fine-tuning of generative models. Unfortunately, PPO requires multiple heuristics to enable stable convergence (e.g. value networks, clipping), and is notorious for its sensitivity to the precise implementation of these components. In response, we take a step back and ask what a minimalist RL algorithm for the era of generative models would look like. We propose REBEL, an algorithm that cleanly reduces the problem of policy optimization to regressing the relative reward between two completions to a prompt in terms of the policy, enabling strikingly lightweight implementation. In theory, we prove that fundamental RL algorithms like Natural Policy Gradient can be seen as variants of REBEL, which allows us to match the strongest known theoretical guarantees in terms of convergence and sample complexity in the RL literature. REBEL can also cleanly incorporate offline data and be extended to handle the intransitive preferences we frequently see in practice. Empirically, we find that REBEL provides a unified approach to language modeling and image generation with stronger or similar performance as PPO and DPO, all while being simpler to implement and more computationally efficient than PPO. When fine-tuning Llama-3-8B-Instruct, REBEL achieves strong performance in AlpacaEval 2.0, MT-Bench, and Open LLM Leaderboard.

Jonathan D. Chang, Dhruv Sreenivas, Yingbing Huang et al.

Adversarial imitation learning (AIL) has stood out as a dominant framework across various imitation learning (IL) applications, with Discriminator Actor Critic (DAC) (Kostrikov et al.,, 2019) demonstrating the effectiveness of off-policy learning algorithms in improving sample efficiency and scalability to higher-dimensional observations. Despite DAC's empirical success, the original AIL objective is on-policy and DAC's ad-hoc application of off-policy training does not guarantee successful imitation (Kostrikov et al., 2019; 2020). Follow-up work such as ValueDICE (Kostrikov et al., 2020) tackles this issue by deriving a fully off-policy AIL objective. Instead in this work, we develop a novel and principled AIL algorithm via the framework of boosting. Like boosting, our new algorithm, AILBoost, maintains an ensemble of properly weighted weak learners (i.e., policies) and trains a discriminator that witnesses the maximum discrepancy between the distributions of the ensemble and the expert policy. We maintain a weighted replay buffer to represent the state-action distribution induced by the ensemble, allowing us to train discriminators using the entire data collected so far. In the weighted replay buffer, the contribution of the data from older policies are properly discounted with the weight computed based on the boosting framework. Empirically, we evaluate our algorithm on both controller state-based and pixel-based environments from the DeepMind Control Suite. AILBoost outperforms DAC on both types of environments, demonstrating the benefit of properly weighting replay buffer data for off-policy training. On state-based environments, DAC outperforms ValueDICE and IQ-Learn (Gary et al., 2021), achieving competitive performance with as little as one expert trajectory.

Jonathan Chang, Nishanth Kumar, Sean Hastings et al.

Robots need to learn skills that can not only generalize across similar problems but also be directed to a specific goal. Previous methods either train a new skill for every different goal or do not infer the specific target in the presence of multiple goals from visual data. We introduce an end-to-end method that represents targetable visuomotor skills as a goal-parameterized neural network policy. By training on an informative subset of available goals with the associated target parameters, we are able to learn a policy that can zero-shot generalize to previously unseen goals. We evaluate our method in a representative 2D simulation of a button-grid and on both button-pressing and peg-insertion tasks on two different physical arms. We demonstrate that our model trained on 33% of the possible goals is able to generalize to more than 90% of the targets in the scene for both simulation and robot experiments. We also successfully learn a mapping from target pixel coordinates to a robot policy to complete a specified goal.

Deepayan Chakrabarti, Stanislav Funiak, Jonathan Chang et al.

We tackle the problem of inferring node labels in a partially labeled graph where each node in the graph has multiple label types and each label type has a large number of possible labels. Our primary example, and the focus of this paper, is the joint inference of label types such as hometown, current city, and employers, for users connected by a social network. Standard label propagation fails to consider the properties of the label types and the interactions between them. Our proposed method, called EdgeExplain, explicitly models these, while still enabling scalable inference under a distributed message-passing architecture. On a billion-node subset of the Facebook social network, EdgeExplain significantly outperforms label propagation for several label types, with lifts of up to 120% for recall@1 and 60% for recall@3.