While Reinforcement Learning from Human Feedback (RLHF) aligns Large Language Models (LLMs) with general, aggregate human preferences, it is suboptimal for learning diverse, individual perspectives. In this work, we study Reinforcement Learning from Personalized Human Feedback (RLPHF) problem, wherein LLMs are aligned to multiple (sometimes conflicting) preferences by modeling alignment as a Multi-Objective Reinforcement Learning (MORL) problem. Compared to strong single-objective baselines, we show that we can achieve personalized alignment by decomposing preferences into multiple dimensions. These dimensions are defined based on personalizations that are declared as desirable by the user. In this work, we show that they can be efficiently trained independently in a distributed manner and combined effectively post-hoc through parameter merging. The code is available at https://github.com/joeljang/RLPHF.
Reinforcement Learning from Human Feedback (RLHF) has shown potential in qualitative tasks where easily defined performance measures are lacking. However, there are drawbacks when RLHF is commonly used to optimize for average human preferences, especially in generative tasks that demand diverse model responses. Meanwhile, Quality Diversity (QD) algorithms excel at identifying diverse and high-quality solutions but often rely on manually crafted diversity metrics. This paper introduces Quality Diversity through Human Feedback (QDHF), a novel approach that progressively infers diversity metrics from human judgments of similarity among solutions, thereby enhancing the applicability and effectiveness of QD algorithms in complex and open-ended domains. Empirical studies show that QDHF significantly outperforms state-of-the-art methods in automatic diversity discovery and matches the efficacy of QD with manually crafted diversity metrics on standard benchmarks in robotics and reinforcement learning. Notably, in open-ended generative tasks, QDHF substantially enhances the diversity of text-to-image generation from a diffusion model and is more favorably received in user studies. We conclude by analyzing QDHF's scalability, robustness, and quality of derived diversity metrics, emphasizing its strength in open-ended optimization tasks. Code and tutorials are available at https://liding.info/qdhf.
Maintaining a consistent persona is a key quality for any open domain dialogue system. Current state-of-the-art systems do this by training agents with supervised learning or online reinforcement learning (RL). However, systems trained with supervised learning often lack consistency as they are never punished for uttering contradictions. Additional training with RL can alleviate some of these issues, however the training process is expensive. Instead, we propose an offline RL framework to improve the persona consistency of dialogue systems. Our framework allows us to combine the advantages of previous methods as we can inexpensively train our model on existing data as in supervised learning, while punishing and rewarding specific utterances as in RL. We also introduce a simple importance sampling method to reduce the variance of importance weights in offline RL training which we call Variance-Reducing MLE-Initialized (VaRMI) importance sampling. Our automatic and human evaluations show that our framework improves both the persona consistency and dialogue quality of a state-of-the-art social chatbot.
Envisioned application areas for reinforcement learning (RL) include autonomous driving, precision agriculture, and finance, which all require RL agents to make decisions in the real world. A significant challenge hindering the adoption of RL methods in these domains is the non-robustness of conventional algorithms. In particular, the focus of RL is typically on the expected value of the return. The expected value is the average over the statistical ensemble of infinitely many trajectories, which can be uninformative about the performance of the average individual. For instance, when we have a heavy-tailed return distribution, the ensemble average can be dominated by rare extreme events. Consequently, optimizing the expected value can lead to policies that yield exceptionally high returns with a probability that approaches zero but almost surely result in catastrophic outcomes in single long trajectories. In this paper, we develop an algorithm that lets RL agents optimize the long-term performance of individual trajectories. The algorithm enables the agents to learn robust policies, which we show in an instructive example with a heavy-tailed return distribution and standard RL benchmarks. The key element of the algorithm is a transformation that we learn from data. This transformation turns the time series of collected returns into one for whose increments expected value and the average over a long trajectory coincide. Optimizing these increments results in robust policies.
We introduce AMAGO, an in-context Reinforcement Learning (RL) agent that uses sequence models to tackle the challenges of generalization, long-term memory, and meta-learning. Recent works have shown that off-policy learning can make in-context RL with recurrent policies viable. Nonetheless, these approaches require extensive tuning and limit scalability by creating key bottlenecks in agents' memory capacity, planning horizon, and model size. AMAGO revisits and redesigns the off-policy in-context approach to successfully train long-sequence Transformers over entire rollouts in parallel with end-to-end RL. Our agent is scalable and applicable to a wide range of problems, and we demonstrate its strong performance empirically in meta-RL and long-term memory domains. AMAGO's focus on sparse rewards and off-policy data also allows in-context learning to extend to goal-conditioned problems with challenging exploration. When combined with a multi-goal hindsight relabeling scheme, AMAGO can solve a previously difficult category of open-world domains, where agents complete many possible instructions in procedurally generated environments.
Visual active search (VAS) has been proposed as a modeling framework in which visual cues are used to guide exploration, with the goal of identifying regions of interest in a large geospatial area. Its potential applications include identifying hot spots of rare wildlife poaching activity, search-and-rescue scenarios, identifying illegal trafficking of weapons, drugs, or people, and many others. State of the art approaches to VAS include applications of deep reinforcement learning (DRL), which yield end-to-end search policies, and traditional active search, which combines predictions with custom algorithmic approaches. While the DRL framework has been shown to greatly outperform traditional active search in such domains, its end-to-end nature does not make full use of supervised information attained either during training, or during actual search, a significant limitation if search tasks differ significantly from those in the training distribution. We propose an approach that combines the strength of both DRL and conventional active search by decomposing the search policy into a prediction module, which produces a geospatial distribution of regions of interest based on task embedding and search history, and a search module, which takes the predictions and search history as input and outputs the search distribution. We develop a novel meta-learning approach for jointly learning the resulting combined policy that can make effective use of supervised information obtained both at training and decision time. Our extensive experiments demonstrate that the proposed representation and meta-learning frameworks significantly outperform state of the art in visual active search on several problem domains.
Machine learning is a rapidly growing field with the potential to revolutionize many areas of science, including physics. This review provides a brief overview of machine learning in physics, covering the main concepts of supervised, unsupervised, and reinforcement learning, as well as more specialized topics such as causal inference, symbolic regression, and deep learning. We present some of the principal applications of machine learning in physics and discuss the associated challenges and perspectives.
A critical goal of autonomy and artificial intelligence is enabling autonomous robots to rapidly adapt in dynamic and uncertain environments. Classic adaptive control and safe control provide stability and safety guarantees but are limited to specific system classes. In contrast, policy adaptation based on reinforcement learning (RL) offers versatility and generalizability but presents safety and robustness challenges. We propose SafeDPA, a novel RL and control framework that simultaneously tackles the problems of policy adaptation and safe reinforcement learning. SafeDPA jointly learns adaptive policy and dynamics models in simulation, predicts environment configurations, and fine-tunes dynamics models with few-shot real-world data. A safety filter based on the Control Barrier Function (CBF) on top of the RL policy is introduced to ensure safety during real-world deployment. We provide theoretical safety guarantees of SafeDPA and show the robustness of SafeDPA against learning errors and extra perturbations. Comprehensive experiments on (1) classic control problems (Inverted Pendulum), (2) simulation benchmarks (Safety Gym), and (3) a real-world agile robotics platform (RC Car) demonstrate great superiority of SafeDPA in both safety and task performance, over state-of-the-art baselines. Particularly, SafeDPA demonstrates notable generalizability, achieving a 300% increase in safety rate compared to the baselines, under unseen disturbances in real-world experiments.
Reinforcement Learning (RL) has achieved significant milestones in the gaming domain, most notably Google DeepMind's AlphaGo defeating human Go champion Ken Jie. This victory was also made possible through the Atari Learning Environment (ALE): The ALE has been foundational in RL research, facilitating significant RL algorithm developments such as AlphaGo and others. In current Atari video game RL research, RL agents' perceptions of its environment is based on raw pixel data from the Atari video game screen with minimal image preprocessing. Contrarily, cutting-edge ML research, external to the Atari video game RL research domain, is focusing on enhancing image perception. A notable example is Meta Research's "Segment Anything Model" (SAM), a foundation model capable of segmenting images without prior training (zero-shot). This paper addresses a novel methodical question: Can state-of-the-art image segmentation models such as SAM improve the performance of RL agents playing Atari video games? The results suggest that SAM can serve as a "virtual augmented reality" for the RL agent, boosting its Atari video game playing performance under certain conditions. Comparing RL agent performance results from raw and augmented pixel inputs provides insight into these conditions. Although this paper was limited by computational constraints, the findings show improved RL agent performance for augmented pixel inputs and can inform broader research agendas in the domain of "virtual augmented reality for video game playing RL agents".
Large transformer models pretrained on offline reinforcement learning datasets have demonstrated remarkable in-context reinforcement learning (ICRL) capabilities, where they can make good decisions when prompted with interaction trajectories from unseen environments. However, when and how transformers can be trained to perform ICRL have not been theoretically well-understood. In particular, it is unclear which reinforcement-learning algorithms transformers can perform in context, and how distribution mismatch in offline training data affects the learned algorithms. This paper provides a theoretical framework that analyzes supervised pretraining for ICRL. This includes two recently proposed training methods -- algorithm distillation and decision-pretrained transformers. First, assuming model realizability, we prove the supervised-pretrained transformer will imitate the conditional expectation of the expert algorithm given the observed trajectory. The generalization error will scale with model capacity and a distribution divergence factor between the expert and offline algorithms. Second, we show transformers with ReLU attention can efficiently approximate near-optimal online reinforcement learning algorithms like LinUCB and Thompson sampling for stochastic linear bandits, and UCB-VI for tabular Markov decision processes. This provides the first quantitative analysis of the ICRL capabilities of transformers pretrained from offline trajectories.
Process synthesis in chemical engineering is a complex planning problem due to vast search spaces, continuous parameters and the need for generalization. Deep reinforcement learning agents, trained without prior knowledge, have shown to outperform humans in various complex planning problems in recent years. Existing work on reinforcement learning for flowsheet synthesis shows promising concepts, but focuses on narrow problems in a single chemical system, limiting its practicality. We present a general deep reinforcement learning approach for flowsheet synthesis. We demonstrate the adaptability of a single agent to the general task of separating binary azeotropic mixtures. Without prior knowledge, it learns to craft near-optimal flowsheets for multiple chemical systems, considering different feed compositions and conceptual approaches. On average, the agent can separate more than 99% of the involved materials into pure components, while autonomously learning fundamental process engineering paradigms. This highlights the agent's planning flexibility, an encouraging step toward true generality.
Due to its training stability and strong expression, the diffusion model has attracted considerable attention in offline reinforcement learning. However, several challenges have also come with it: 1) The demand for a large number of diffusion steps makes the diffusion-model-based methods time inefficient and limits their applications in real-time control; 2) How to achieve policy improvement with accurate guidance for diffusion model-based policy is still an open problem. Inspired by the consistency model, we propose a novel time-efficiency method named Consistency Policy with Q-Learning (CPQL), which derives action from noise by a single step. By establishing a mapping from the reverse diffusion trajectories to the desired policy, we simultaneously address the issues of time efficiency and inaccurate guidance when updating diffusion model-based policy with the learned Q-function. We demonstrate that CPQL can achieve policy improvement with accurate guidance for offline reinforcement learning, and can be seamlessly extended for online RL tasks. Experimental results indicate that CPQL achieves new state-of-the-art performance on 11 offline and 21 online tasks, significantly improving inference speed by nearly 45 times compared to Diffusion-QL. We will release our code later.
Recent advances in robot-assisted surgery have resulted in progressively more precise, efficient, and minimally invasive procedures, sparking a new era of robotic surgical intervention. This enables doctors, in collaborative interaction with robots, to perform traditional or minimally invasive surgeries with improved outcomes through smaller incisions. Recent efforts are working toward making robotic surgery more autonomous which has the potential to reduce variability of surgical outcomes and reduce complication rates. Deep reinforcement learning methodologies offer scalable solutions for surgical automation, but their effectiveness relies on extensive data acquisition due to the absence of prior knowledge in successfully accomplishing tasks. Due to the intensive nature of simulated data collection, previous works have focused on making existing algorithms more efficient. In this work, we focus on making the simulator more efficient, making training data much more accessible than previously possible. We introduce Surgical Gym, an open-source high performance platform for surgical robot learning where both the physics simulation and reinforcement learning occur directly on the GPU. We demonstrate between 100-5000x faster training times compared with previous surgical learning platforms. The code is available at: https://github.com/SamuelSchmidgall/SurgicalGym.
This investigation introduces a novel deep reinforcement learning-based suite to control floating platforms in both simulated and real-world environments. Floating platforms serve as versatile test-beds to emulate micro-gravity environments on Earth, useful to test autonomous navigation systems for space applications. Our approach addresses the system and environmental uncertainties in controlling such platforms by training policies capable of precise maneuvers amid dynamic and unpredictable conditions. Leveraging Deep Reinforcement Learning (DRL) techniques, our suite achieves robustness, adaptability, and good transferability from simulation to reality. Our deep reinforcement learning framework provides advantages such as fast training times, large-scale testing capabilities, rich visualization options, and ROS bindings for integration with real-world robotic systems. Being open access, our suite serves as a comprehensive platform for practitioners who want to replicate similar research in their own simulated environments and labs.
The performance of existing supervised neuron segmentation methods is highly dependent on the number of accurate annotations, especially when applied to large scale electron microscopy (EM) data. By extracting semantic information from unlabeled data, self-supervised methods can improve the performance of downstream tasks, among which the mask image model (MIM) has been widely used due to its simplicity and effectiveness in recovering original information from masked images. However, due to the high degree of structural locality in EM images, as well as the existence of considerable noise, many voxels contain little discriminative information, making MIM pretraining inefficient on the neuron segmentation task. To overcome this challenge, we propose a decision-based MIM that utilizes reinforcement learning (RL) to automatically search for optimal image masking ratio and masking strategy. Due to the vast exploration space, using single-agent RL for voxel prediction is impractical. Therefore, we treat each input patch as an agent with a shared behavior policy, allowing for multi-agent collaboration. Furthermore, this multi-agent model can capture dependencies between voxels, which is beneficial for the downstream segmentation task. Experiments conducted on representative EM datasets demonstrate that our approach has a significant advantage over alternative self-supervised methods on the task of neuron segmentation. Code is available at \url{https://github.com/ydchen0806/dbMiM}.
In this paper, a unified framework for exploration in reinforcement learning (RL) is proposed based on an option-critic model. The proposed framework learns to integrate a set of diverse exploration strategies so that the agent can adaptively select the most effective exploration strategy over time to realize a relevant exploration-exploitation trade-off for each given task. The effectiveness of the proposed exploration framework is demonstrated by various experiments in the MiniGrid and Atari environments.
Reinforcement learning (RL) for complex tasks remains a challenge, primarily due to the difficulties of engineering scalar reward functions and the inherent inefficiency of training models from scratch. Instead, it would be better to specify complex tasks in terms of elementary subtasks and to reuse subtask solutions whenever possible. In this work, we address continuous space lexicographic multi-objective RL problems, consisting of prioritized subtasks, which are notoriously difficult to solve. We show that these can be scalarized with a subtask transformation and then solved incrementally using value decomposition. Exploiting this insight, we propose prioritized soft Q-decomposition (PSQD), a novel algorithm for learning and adapting subtask solutions under lexicographic priorities in continuous state-action spaces. PSQD offers the ability to reuse previously learned subtask solutions in a zero-shot composition, followed by an adaptation step. Its ability to use retained subtask training data for offline learning eliminates the need for new environment interaction during adaptation. We demonstrate the efficacy of our approach by presenting successful learning, reuse, and adaptation results for both low- and high-dimensional simulated robot control tasks, as well as offline learning results. In contrast to baseline approaches, PSQD does not trade off between conflicting subtasks or priority constraints and satisfies subtask priorities during learning. PSQD provides an intuitive framework for tackling complex RL problems, offering insights into the inner workings of the subtask composition.
Recent challenges in operating power networks arise from increasing energy demands and unpredictable renewable sources like wind and solar. While reinforcement learning (RL) shows promise in managing these networks, through topological actions like bus and line switching, efficiently handling large action spaces as networks grow is crucial. This paper presents a hierarchical multi-agent reinforcement learning (MARL) framework tailored for these expansive action spaces, leveraging the power grid's inherent hierarchical nature. Experimental results indicate the MARL framework's competitive performance with single-agent RL methods. We also compare different RL algorithms for lower-level agents alongside different policies for higher-order agents.
Perceptive deep reinforcement learning (DRL) has lead to many recent breakthroughs for complex AI systems leveraging image-based input data. Applications of these results range from super-human level video game agents to dexterous, physically intelligent robots. However, training these perceptive DRL-enabled systems remains incredibly compute and memory intensive, often requiring huge training datasets and large experience replay buffers. This poses a challenge for the next generation of field robots that will need to be able to learn on the edge in order to adapt to their environments. In this paper, we begin to address this issue through differentially encoded observation spaces. By reinterpreting stored image-based observations as a video, we leverage lossless differential video encoding schemes to compress the replay buffer without impacting training performance. We evaluate our approach with three state-of-the-art DRL algorithms and find that differential image encoding reduces the memory footprint by as much as 14.2x and 16.7x across tasks from the Atari 2600 benchmark and the DeepMind Control Suite (DMC) respectively. These savings also enable large-scale perceptive DRL that previously required paging between flash and RAM to be run entirely in RAM, improving the latency of DMC tasks by as much as 32%.
We develop a general theory to optimize the frequentist regret for sequential learning problems, where efficient bandit and reinforcement learning algorithms can be derived from unified Bayesian principles. We propose a novel optimization approach to generate "algorithmic beliefs" at each round, and use Bayesian posteriors to make decisions. The optimization objective to create "algorithmic beliefs," which we term "Algorithmic Information Ratio," represents an intrinsic complexity measure that effectively characterizes the frequentist regret of any algorithm. To the best of our knowledge, this is the first systematical approach to make Bayesian-type algorithms prior-free and applicable to adversarial settings, in a generic and optimal manner. Moreover, the algorithms are simple and often efficient to implement. As a major application, we present a novel algorithm for multi-armed bandits that achieves the "best-of-all-worlds" empirical performance in the stochastic, adversarial, and non-stationary environments. And we illustrate how these principles can be used in linear bandits, bandit convex optimization, and reinforcement learning.