We study offline reinforcement learning in average-reward MDPs, which presents increased challenges from the perspectives of distribution shift and non-uniform coverage, and has been relatively underexamined from a theoretical perspective. While previous work obtains performance guarantees under single-policy data coverage assumptions, such guarantees utilize additional complexity measures which are uniform over all policies, such as the uniform mixing time. We develop sharp guarantees depending only on the target policy, specifically the bias span and a novel policy hitting radius, yielding the first fully single-policy sample complexity bound for average-reward offline RL. We are also the first to handle general weakly communicating MDPs, contrasting restrictive structural assumptions made in prior work. To achieve this, we introduce an algorithm based on pessimistic discounted value iteration enhanced by a novel quantile clipping technique, which enables the use of a sharper empirical-span-based penalty function. Our algorithm also does not require any prior parameter knowledge for its implementation. Remarkably, we show via hard examples that learning under our conditions requires coverage assumptions beyond the stationary distribution of the target policy, distinguishing single-policy complexity measures from previously examined cases. We also develop lower bounds nearly matching our main result.

This paper introduces a new approach for approximating the learning dynamics of multiple reinforcement learning (RL) agents interacting in a finite-state Markov game. The idea is to rescale the learning process by simultaneously reducing the learning rate and increasing the update frequency, effectively treating the agent's parameters as a slow-evolving variable influenced by the fast-mixing game state. Under mild assumptions-ergodicity of the state process and continuity of the updates-we prove the convergence of this rescaled process to an ordinary differential equation (ODE). This ODE provides a tractable, deterministic approximation of the agent's learning dynamics. An implementation of the framework is available at\,: https://github.com/yannKerzreho/MarkovGameApproximation

Adapting a pretrained diffusion model to new objectives at inference time remains an open problem in generative modeling. Existing steering methods suffer from inaccurate value estimation, especially at high noise levels, which biases guidance. Moreover, information from past runs is not reused to improve sample quality, resulting in inefficient use of compute. Inspired by the success of Monte Carlo Tree Search, we address these limitations by casting inference-time alignment as a search problem that reuses past computations. We introduce a tree-based approach that samples from the reward-aligned target density by propagating terminal rewards back through the diffusion chain and iteratively refining value estimates with each additional generation. Our proposed method, Diffusion Tree Sampling (DTS), produces asymptotically exact samples from the target distribution in the limit of infinite rollouts, and its greedy variant, Diffusion Tree Search (DTS$^\star$), performs a global search for high reward samples. On MNIST and CIFAR-10 class-conditional generation, DTS matches the FID of the best-performing baseline with up to $10\times$ less compute. In text-to-image generation and language completion tasks, DTS$^\star$ effectively searches for high reward samples that match best-of-N with up to $5\times$ less compute. By reusing information from previous generations, we get an anytime algorithm that turns additional compute into steadily better samples, providing a scalable approach for inference-time alignment of diffusion models.

We study value-iteration (VI) algorithms for solving general (a.k.a. multichain) Markov decision processes (MDPs) under the average-reward criterion, a fundamental but theoretically challenging setting. Beyond the difficulties inherent to all average-reward problems posed by the lack of contractivity and non-uniqueness of solutions to the Bellman operator, in the multichain setting an optimal policy must solve the navigation subproblem of steering towards the best connected component, in addition to optimizing long-run performance within each component. We develop algorithms which better solve this navigational subproblem in order to achieve faster convergence for multichain MDPs, obtaining improved rates of convergence and sharper measures of complexity relative to prior work. Many key components of our results are of potential independent interest, including novel connections between average-reward and discounted problems, optimal fixed-point methods for discounted VI which extend to general Banach spaces, new sublinear convergence rates for the discounted value error, and refined suboptimality decompositions for multichain MDPs. Overall our results yield faster convergence rates for discounted and average-reward problems and expand the theoretical foundations of VI approaches.

Robotic control policies learned from human demonstrations have achieved impressive results in many real-world applications. However, in scenarios where initial performance is not satisfactory, as is often the case in novel open-world settings, such behavioral cloning (BC)-learned policies typically require collecting additional human demonstrations to further improve their behavior -- an expensive and time-consuming process. In contrast, reinforcement learning (RL) holds the promise of enabling autonomous online policy improvement, but often falls short of achieving this due to the large number of samples it typically requires. In this work we take steps towards enabling fast autonomous adaptation of BC-trained policies via efficient real-world RL. Focusing in particular on diffusion policies -- a state-of-the-art BC methodology -- we propose diffusion steering via reinforcement learning (DSRL): adapting the BC policy by running RL over its latent-noise space. We show that DSRL is highly sample efficient, requires only black-box access to the BC policy, and enables effective real-world autonomous policy improvement. Furthermore, DSRL avoids many of the challenges associated with finetuning diffusion policies, obviating the need to modify the weights of the base policy at all. We demonstrate DSRL on simulated benchmarks, real-world robotic tasks, and for adapting pretrained generalist policies, illustrating its sample efficiency and effective performance at real-world policy improvement.

We propose a framework which generalizes "decision making with structured observations" by allowing robust (i.e. multivalued) models. In this framework, each model associates each decision with a convex set of probability distributions over outcomes. Nature can choose distributions out of this set in an arbitrary (adversarial) manner, that can be nonoblivious and depend on past history. The resulting framework offers much greater generality than classical bandits and reinforcement learning, since the realizability assumption becomes much weaker and more realistic. We then derive a theory of regret bounds for this framework. Although our lower and upper bounds are not tight, they are sufficient to fully characterize power-law learnability. We demonstrate this theory in two special cases: robust linear bandits and tabular robust online reinforcement learning. In both cases, we derive regret bounds that improve state-of-the-art (except that we do not address computational efficiency).

Long-horizon goal-conditioned tasks pose fundamental challenges for reinforcement learning (RL), particularly when goals are distant and rewards are sparse. While hierarchical and graph-based methods offer partial solutions, they often suffer from subgoal infeasibility and inefficient planning. We introduce Strict Subgoal Execution (SSE), a graph-based hierarchical RL framework that enforces single-step subgoal reachability by structurally constraining high-level decision-making. To enhance exploration, SSE employs a decoupled exploration policy that systematically traverses underexplored regions of the goal space. Furthermore, a failure-aware path refinement, which refines graph-based planning by dynamically adjusting edge costs according to observed low-level success rates, thereby improving subgoal reliability. Experimental results across diverse long-horizon benchmarks demonstrate that SSE consistently outperforms existing goal-conditioned RL and hierarchical RL approaches in both efficiency and success rate.

Modern deep architectures often rely on large-scale datasets, but training on these datasets incurs high computational and storage overhead. Real-world datasets often contain substantial redundancies, prompting the need for more data-efficient training paradigms. Data selection has shown promise to mitigate redundancy by identifying the most representative samples, thereby reducing training costs without compromising performance. Existing methods typically rely on static scoring metrics or pretrained models, overlooking the combined effect of selected samples and their evolving dynamics during training. We introduce the concept of epsilon-sample cover, which quantifies sample redundancy based on inter-sample relationships, capturing the intrinsic structure of the dataset. Based on this, we reformulate data selection as a reinforcement learning (RL) process and propose RL-Selector, where a lightweight RL agent optimizes the selection policy by leveraging epsilon-sample cover derived from evolving dataset distribution as a reward signal. Extensive experiments across benchmark datasets and diverse architectures demonstrate that our method consistently outperforms existing state-of-the-art baselines. Models trained with our selected datasets show enhanced generalization performance with improved training efficiency.

--The time allocation problem in multi-function cognitive radar systems focuses on the trade-off between scanning for newly emerging targets and tracking the previously detected targets. We formulate this as a multi-objective optimization problem and employ deep reinforcement learning to find Pareto-optimal solutions and compare deep deterministic policy gradient (DDPG) and soft actor-critic (SAC) algorithms. Our results demonstrate the effectiveness of both algorithms in adapting to various scenarios, with SAC showing improved stability and sample efficiency compared to DDPG. We further employ the NSGA-II algorithm to estimate an upper bound on the Pareto front of the considered problem. This work contributes to the development of more efficient and adaptive cognitive radar systems capable of balancing multiple competing objectives in dynamic environments.

-- In this paper, we tackle the task of adaptive time allocation in integrated sensing and communication systems equipped with radar and communication units. The dual-functional radar-communication system's task involves allocating dwell times for tracking multiple targets and utilizing the remaining time for data transmission towards estimated target locations. We introduce a novel constrained deep reinforcement learning (CDRL) approach, designed to optimize resource allocation between tracking and communication under time budget constraints, thereby enhancing target communication quality. Our numerical results demonstrate the efficiency of our proposed CDRL framework, confirming its ability to maximize communication quality in highly dynamic environments while adhering to time constraints. A. Background 1) Cognitive Radar: Radar technology, integral to various applications in environmental sensing, space exploration, navigation, and traffic control, has become increasingly important with the emergence of autonomous vehicles and drones.