Recent advances in reinforcement learning have shown that language models can develop sophisticated reasoning through training on tasks with verifiable rewards, but these approaches depend on human-curated problem-answer pairs and domain-specific reward engineering. We introduce SPIRAL, a self-play framework where models learn by playing multi-turn, zero-sum games against continuously improving versions of themselves, eliminating the need for human supervision. Through self-play, SPIRAL generates an infinite curriculum of progressively challenging problems as models must constantly adapt to stronger opponents. To enable this self-play training at scale, We implement a fully online, multi-turn, multi-agent reinforcement learning system for LLMs and propose role-conditioned advantage estimation (RAE) to stabilize multi-agent training. Using SPIRAL, self-play on zero-sum games produces reasoning capabilities that transfer broadly. Training Qwen3-4B-Base on Kuhn Poker alone achieves 8.6% improvement on math and 8.4% on general reasoning, outperforming SFT on 25,000 expert game trajectories. Analysis reveals that this transfer occurs through three cognitive patterns: systematic decomposition, expected value calculation, and case-by-case analysis. Multi-game training (TicTacToe, Kuhn Poker, Simple Negotiation) further enhances performance as each game develops distinct reasoning strengths. Applying SPIRAL to a strong reasoning model (DeepSeek-R1-Distill-Qwen-7B) can still lead to 2.0% average improvement. These results demonstrate that zero-sum games naturally develop transferable reasoning capabilities, highlighting a promising direction for autonomous reasoning development.

Recent work uses reinforcement learning (RL) to fine-tune text-to-image diffusion models, improving text-image alignment and sample quality. However, existing approaches introduce unnecessary complexity: they cache the full sampling trajectory, depend on differentiable reward models or large preference datasets, or require specialized guidance techniques. Motivated by the "golden noise" hypothesis -- that certain initial noise samples can consistently yield superior alignment -- we introduce Noise PPO, a minimalist RL algorithm that leaves the pre-trained diffusion model entirely frozen and learns a prompt-conditioned initial noise generator. Our approach requires no trajectory storage, reward backpropagation, or complex guidance tricks. Extensive experiments show that optimizing the initial noise distribution consistently improves alignment and sample quality over the original model, with the most significant gains at low inference steps. As the number of inference steps increases, the benefit of noise optimization diminishes but remains present. These findings clarify the scope and limitations of the golden noise hypothesis and reinforce the practical value of minimalist RL fine-tuning for diffusion models.

This paper introduces von Mises-Fisher exploration (vMF-exp), a scalable method for exploring large action sets in reinforcement learning problems where hyperspherical embedding vectors represent these actions. vMF-exp involves initially sampling a state embedding representation using a von Mises-Fisher distribution, then exploring this representation's nearest neighbors, which scales to virtually unlimited numbers of candidate actions. We show that, under theoretical assumptions, vMF-exp asymptotically maintains the same probability of exploring each action as Boltzmann Exploration (B-exp), a popular alternative that, nonetheless, suffers from scalability issues as it requires computing softmax values for each action. Consequently, vMF-exp serves as a scalable alternative to B-exp for exploring large action sets with hyperspherical embeddings. Experiments on simulated data, real-world public data, and the successful large-scale deployment of vMF-exp on the recommender system of a global music streaming service empirically validate the key properties of the proposed method.

Reinforcement Learning (RL) is widely used in tasks where agents interact with an environment to maximize rewards. Building on this foundation, Safe Reinforcement Learning (Safe RL) incorporates a cost metric alongside the reward metric, ensuring that agents adhere to safety constraints during decision-making. In this paper, we identify that Safe RL is vulnerable to backdoor attacks, which can manipulate agents into performing unsafe actions. First, we introduce the relevant concepts and evaluation metrics for backdoor attacks in Safe RL. It is the first attack framework in the Safe RL field that involves both Positive and Negative Action sample (PNAct) is to implant backdoors, where positive action samples provide reference actions and negative action samples indicate actions to be avoided. We theoretically point out the properties of PNAct and design an attack algorithm. Finally, we conduct experiments to evaluate the effectiveness of our proposed backdoor attack framework, evaluating it with the established metrics. This paper highlights the potential risks associated with Safe RL and underscores the feasibility of such attacks. Our code and supplementary material are available at https://github.com/azure-123/PNAct.

Deep Reinforcement Learning (DRL) has achieved remarkable success in complex sequential decision-making tasks, such as playing Atari 2600 games and mastering board games. A critical yet underexplored aspect of DRL is the temporal scale of action execution. We propose a novel paradigm that integrates contextual bandits with DRL to adaptively select action durations, enhancing policy flexibility and computational efficiency. Our approach augments a Deep Q-Network (DQN) with a contextual bandit module that learns to choose optimal action repetition rates based on state contexts. Experiments on Atari 2600 games demonstrate significant performance improvements over static duration baselines, highlighting the efficacy of adaptive temporal abstractions in DRL. This paradigm offers a scalable solution for real-time applications like gaming and robotics, where dynamic action durations are critical.

With the growing demand for novel materials, machine learning-driven inverse design methods face significant challenges in reconciling the high-dimensional materials composition space with limited experimental data. Existing approaches suffer from two major limitations: (I) machine learning models often lack reliability in high-dimensional spaces, leading to prediction biases during the design process; (II) these models fail to effectively incorporate domain expert knowledge, limiting their capacity to support knowledge-guided inverse design. To address these challenges, we introduce AIMatDesign, a reinforcement learning framework that addresses these limitations by augmenting experimental data using difference-based algorithms to build a trusted experience pool, accelerating model convergence. To enhance model reliability, an automated refinement strategy guided by large language models (LLMs) dynamically corrects prediction inconsistencies, reinforcing alignment between reward signals and state value functions. Additionally, a knowledge-based reward function leverages expert domain rules to improve stability and efficiency during training. Our experiments demonstrate that AIMatDesign significantly surpasses traditional machine learning and reinforcement learning methods in discovery efficiency, convergence speed, and success rates. Among the numerous candidates proposed by AIMatDesign, experimental synthesis of representative Zr-based alloys yielded a top-performing BMG with 1.7GPa yield strength and 10.2\% elongation, closely matching predictions. Moreover, the framework accurately captured the trend of yield strength variation with composition, demonstrating its reliability and potential for closed-loop materials discovery.

--In this work we revisit the Mobility Robustness Optimisation (MRO) algorithm and study the possibility of learning the optimal Cell Individual Offset tuning using offline Reinforcement Learning. Such methods make use of collected offline datasets to learn the optimal policy, without further exploration. We adapt and apply a sequence-based method called Decision Transformers as well as a value-based method called Conservative Q-Learning to learn the optimal policy for the same target reward as the vanilla rule-based MRO. The same input features related to failures, ping-pongs, and other handover issues are used. Evaluation for realistic New Radio networks with 3500 MHz carrier frequency on a traffic mix including diverse user service types and a specific tunable cell-pair shows that offline-RL methods outperform rule-based MRO, offering up to 7% improvement. Furthermore, offline-RL can be trained for diverse objective functions using the same available dataset, thus offering operational flexibility compared to rule-based methods. Self-Organizing Network (SON) functionalities have become key aspects of modern cellular networks for automation. They make use of collected data to allow the network to self-configure, self-optimize, and self-heal. Mobility Robustness Optimization (MRO) is the related SON feature whose aim is to optimize the configuration of relevant mobility parameters and allow users to experience seamless connectivity.

Peer-to-peer (P2P) trading is increasingly recognized as a key mechanism for decentralized market regulation, yet existing approaches often lack robust frameworks to ensure fairness. This paper presents FairMarket-RL, a novel hybrid framework that combines Large Language Models (LLMs) with Reinforcement Learning (RL) to enable fairness-aware trading agents. In a simulated P2P microgrid with multiple sellers and buyers, the LLM acts as a real-time fairness critic, evaluating each trading episode using two metrics: Fairness-To-Buyer (FTB) and Fairness-Between-Sellers (FBS). These fairness scores are integrated into agent rewards through scheduled λ-coefficients, forming an adaptive LLM-guided reward shaping loop that replaces brittle, rule-based fairness constraints. Agents are trained using Independent Proximal Policy Optimization (IPPO) and achieve equitable outcomes, fulfilling over 90% of buyer demand, maintaining fair seller margins, and consistently reaching FTB and FBS scores above 0.80. The training process demonstrates that fairness feedback improves convergence, reduces buyer shortfalls, and narrows profit disparities between sellers. With its language-based critic, the framework scales naturally, and its extension to a large power distribution system with household prosumers illustrates its practical applicability. FairMarket-RL thus offers a scalable, equity-driven solution for autonomous trading in decentralized energy systems.

Developing decision-making algorithms for highly automated driving systems remains challenging, since these systems have to operate safely in an open and complex environments. Reinforcement Learning (RL) approaches can learn comprehensive decision policies directly from experience and already show promising results in simple driving tasks. However, current approaches fail to achieve generalizability for more complex driving tasks and lack learning efficiency. Therefore, we present Scenario-based Automated Driving Reinforcement Learning (SAD-RL), the first framework that integrates Reinforcement Learning (RL) of hierarchical policy in a scenario-based environment. A high-level policy selects maneuver templates that are evaluated and executed by a low-level control logic. The scenario-based environment allows to control the training experience for the agent and to explicitly introduce challenging, but rate situations into the training process. Our experiments show that an agent trained using the SAD-RL framework can achieve safe behaviour in easy as well as challenging situations efficiently. Our ablation studies confirmed that both HRL and scenario diversity are essential for achieving these results.

In the domain of algorithmic decision-making, non-Markovian dynamics manifest as a significant impediment, especially for paradigms such as Reinforcement Learning (RL), thereby exerting far-reaching consequences on the advancement and effectiveness of the associated systems. Nevertheless, the existing benchmarks are deficient in comprehensively assessing the capacity of decision algorithms to handle non-Markovian dynamics. To address this deficiency, we have devised a generalized methodology grounded in category theory. Notably, we established the category of Markov Decision Processes (MDP) and the category of non-Markovian Decision Processes (NMDP), and proved the equivalence relationship between them. This theoretical foundation provides a novel perspective for understanding and addressing non-Markovian dynamics. We further introduced non-Markovianity into decision-making problem settings via the History Aggregator for State (HAS). With HAS, we can precisely control the state dependency structure of decision-making problems in the time series. Our analysis demonstrates the effectiveness of our method in representing a broad range of non-Markovian dynamics. This approach facilitates a more rigorous and flexible evaluation of decision algorithms by testing them in problem settings where non-Markovian dynamics are explicitly constructed.