The assumption that data are independent and identically distributed underpins all machine learning. When data are collected sequentially from agent experiences this assumption does not generally hold, as in reinforcement learning. Here, we derive a method that overcomes these limitations by exploiting the statistical mechanics of ergodic processes, which we term maximum diffusion reinforcement learning. By decorrelating agent experiences, our approach provably enables single-shot learning in continuous deployments over the course of individual task attempts. Moreover, we prove our approach generalizes well-known maximum entropy techniques, and robustly exceeds state-of-the-art performance across popular benchmarks. Our results at the nexus of physics, learning, and control pave the way towards more transparent and reliable decision-making in reinforcement learning agents, such as locomoting robots and self-driving cars.

A fundamental challenge of deploying RL to real-world applications is the limited ability to explore or interact with the environment, due to resources, time, or safety constraints. Offline RL, or batch RL, seeks to circumvent this challenge by resorting to history data--which are often collected by executing some possibly unknown behavior policy in the past--with the hope that the history data might already provide significant insights about the targeted optimal policy without further exploration (Levine et al., 2020). Besides maximizing the expected total reward, perhaps an equally important goal--to say the least--for an RL agent is safety and robustness (Garcıa and Fernández, 2015), especially in high-stake applications such as robotics, autonomous driving, clinical trials, financial investments, and so on (Choi et al., 2009; Schulman et al., 2013). It has been observed that a standard RL agent trained in an ideal environment might be extremely sensitive and fail catastrophically when the deployed environment is subject to small adversarial perturbations (Zhang et al., 2020). Consequently, robust RL has attracted a surge of attentions with the goal to learn an optimal policy that is robust to environment perturbations. In fact, providing robustness guarantees becomes even more relevant in the offline setting, which can be formulated as robust offline RL, since the history data is often inevitably collected from a timeframe where it is no longer reasonable to assume model stillness, due to the highly non-stationary and time-varying dynamics of many real-world applications. Altogether, this naturally leads to a question: Can we learn a near-optimal policy which is robust with respect to uncertainties and variabilities of the environments using as few history samples as possible?

Reinforcement Learning (RL) encompasses diverse paradigms, including model-based RL, policy-based RL, and value-based RL, each tailored to approximate the model, optimal policy, and optimal value function, respectively. This work investigates the potential hierarchy of representation complexity -- the complexity of functions to be represented -- among these RL paradigms. We first demonstrate that, for a broad class of Markov decision processes (MDPs), the model can be represented by constant-depth circuits with polynomial size or Multi-Layer Perceptrons (MLPs) with constant layers and polynomial hidden dimension. However, the representation of the optimal policy and optimal value proves to be $\mathsf{NP}$-complete and unattainable by constant-layer MLPs with polynomial size. This demonstrates a significant representation complexity gap between model-based RL and model-free RL, which includes policy-based RL and value-based RL. To further explore the representation complexity hierarchy between policy-based RL and value-based RL, we introduce another general class of MDPs where both the model and optimal policy can be represented by constant-depth circuits with polynomial size or constant-layer MLPs with polynomial size. In contrast, representing the optimal value is $\mathsf{P}$-complete and intractable via a constant-layer MLP with polynomial hidden dimension. This accentuates the intricate representation complexity associated with value-based RL compared to policy-based RL. In summary, we unveil a potential representation complexity hierarchy within RL -- representing the model emerges as the easiest task, followed by the optimal policy, while representing the optimal value function presents the most intricate challenge.

We study how experience with asset price bubbles changes the trading strategies of reinforcement learning (RL) traders and ask whether the change in trading strategies helps to prevent future bubbles. We train the RL traders in a multi-agent market simulation platform, ABIDES, and compare the strategies of traders trained with and without bubble experience. We find that RL traders without bubble experience behave like short-term momentum traders, whereas traders with bubble experience behave like value traders. Therefore, RL traders without bubble experience amplify bubbles, whereas RL traders with bubble experience tend to suppress and sometimes prevent them. This finding suggests that learning from experience is a mechanism for a boom and bust cycle where the experience of a collapsing bubble makes future bubbles less likely for a period of time until the memory fades and bubbles become more likely to form again.

We introduce a novel Proximal Policy Optimization (PPO) algorithm aimed at addressing the challenge of maintaining a uniform proton beam intensity delivery in the Muon to Electron Conversion Experiment (Mu2e) at Fermi National Accelerator Laboratory (Fermilab). Our primary objective is to regulate the spill process to ensure a consistent intensity profile, with the ultimate goal of creating an automated controller capable of providing real-time feedback and calibration of the Spill Regulation System (SRS) parameters on a millisecond timescale. We treat the Mu2e accelerator system as a Markov Decision Process suitable for Reinforcement Learning (RL), utilizing PPO to reduce bias and enhance training stability. A key innovation in our approach is the integration of a neuralized Proportional-Integral-Derivative (PID) controller into the policy function, resulting in a significant improvement in the Spill Duty Factor (SDF) by 13.6%, surpassing the performance of the current PID controller baseline by an additional 1.6%. This paper presents the preliminary offline results based on a differentiable simulator of the Mu2e accelerator. It paves the groundwork for real-time implementations and applications, representing a crucial step towards automated proton beam intensity control for the Mu2e experiment.

Engineering system design, viewed as a decision-making process, faces challenges due to complexity and uncertainty. In this paper, we present a framework proposing the use of the Deep Q-learning algorithm to optimize the design of engineering systems. We outline a step-by-step framework for optimizing engineering system designs. The goal is to find policies that maximize the output of a simulation model given multiple sources of uncertainties. The proposed algorithm handles linear and non-linear multi-stage stochastic problems, where decision variables are discrete, and the objective function and constraints are assessed via a Monte Carlo simulation. We demonstrate the effectiveness of our proposed framework by solving two engineering system design problems in the presence of multiple uncertainties, such as price and demand.

Causal confusion is a phenomenon where an agent learns a policy that reflects imperfect spurious correlations in the data. Such a policy may falsely appear to be optimal during training if most of the training data contain such spurious correlations. This phenomenon is particularly pronounced in domains such as robotics, with potentially large gaps between the open- and closed-loop performance of an agent. In such settings, causally confused models may appear to perform well according to open-loop metrics during training but fail catastrophically when deployed in the real world. In this paper, we study causal confusion in offline reinforcement learning. We investigate whether selectively sampling appropriate points from a dataset of demonstrations may enable offline reinforcement learning agents to disambiguate the underlying causal mechanisms of the environment, alleviate causal confusion in offline reinforcement learning, and produce a safer model for deployment. To answer this question, we consider a set of tailored offline reinforcement learning datasets that exhibit causal ambiguity and assess the ability of active sampling techniques to reduce causal confusion at evaluation. We provide empirical evidence that uniform and active sampling techniques are able to consistently reduce causal confusion as training progresses and that active sampling is able to do so significantly more efficiently than uniform sampling.

Learning policies that can generalize to unseen environments is a fundamental challenge in visual reinforcement learning (RL). While most current methods focus on acquiring robust visual representations through auxiliary supervision, pre-training, or data augmentation, the potential of modern vision foundation models remains underleveraged. In this work, we introduce Segment Anything Model for Generalizable visual RL (SAM-G), a novel framework that leverages the promptable segmentation ability of Segment Anything Model (SAM) to enhance the generalization capabilities of visual RL agents. We utilize image features from DINOv2 and SAM to find correspondence as point prompts to SAM, and then SAM produces high-quality masked images for agents directly. Evaluated across 8 DMControl tasks and 3 Adroit tasks, SAM-G significantly improves the visual generalization ability without altering the RL agents' architecture but merely their observations. Notably, SAM-G achieves 44% and 29% relative improvements on the challenging video hard setting on DMControl and Adroit respectively, compared to state-of-the-art methods. Video and code: https://yanjieze.com/SAM-G/

We study interactive imitation learning, where a learner interactively queries a demonstrating expert for action annotations, aiming to learn a policy that has performance competitive with the expert, using as few annotations as possible. We give an algorithmic framework named Ensemble-based Interactive Imitation Learning (EIIL) that achieves this goal. Theoretically, we prove that an oracle-efficient version of EIIL achieves sharp regret guarantee, given access to samples from some ``explorative'' distribution over states. Empirically, EIIL notably surpasses online and offline imitation learning benchmarks in continuous control tasks. Our work opens up systematic investigations on the benefit of using model ensembles for interactive imitation learning.

Column generation (CG) is one of the most successful approaches for solving large-scale linear programming (LP) problems. Given an LP with a prohibitively large number of variables (i.e., columns), the idea of CG is to explicitly consider only a subset of columns and iteratively add potential columns to improve the objective value. While adding the column with the most negative reduced cost can guarantee the convergence of CG, it has been shown that adding multiple columns per iteration rather than a single column can lead to faster convergence. However, it remains a challenge to design a multiple-column selection strategy to select the most promising columns from a large number of candidate columns. In this paper, we propose a novel reinforcement-learning-based (RL) multiple-column selection strategy. To the best of our knowledge, it is the first RL-based multiple-column selection strategy for CG. The effectiveness of our approach is evaluated on two sets of problems: the cutting stock problem and the graph coloring problem. Compared to several widely used single-column and multiple-column selection strategies, our RL-based multiple-column selection strategy leads to faster convergence and achieves remarkable reductions in the number of CG iterations and runtime.