Model-based reinforcement learning (RL) has shown strong potential in handling these challenges while remaining sample-efficient. Additionally, Dreamer has demonstrated that online model-based RL can be achieved using a recurrent world model trained on replay buffer data. However, applying Dreamer to aerial systems has been quite challenging due to its sample inefficiency and poor generalization of dynamics models. Our work explores a physics-informed approach to world model learning and improves policy performance. The world model treats the quadcopter as a free-body system and predicts the net forces and moments acting on it, which are then passed through a 6-DOF Runge-Kutta integrator (RK4) to predict future state rollouts. In this paper, we compare this physics-informed method to a standard RNN-based world model. Although both models perform well on the training data, we observed that they fail to generalize to new trajectories, leading to rapid divergence in state rollouts, preventing policy convergence.

Intrusion Detection Systems (IDS) play a vital role in defending modern cyber physical systems against increasingly sophisticated cyber threats. Deep Reinforcement Learning-based IDS, have shown promise due to their adaptive and generalization capabilities. However, recent studies reveal their vulnerability to adversarial attacks, including Universal Adversarial Perturbations (UAPs), which can deceive models with a single, input-agnostic perturbation. In this work, we propose a novel UAP attack against Deep Reinforcement Learning (DRL)-based IDS under the domain-specific constraints derived from network data rules and feature relationships. To the best of our knowledge, there is no existing study that has explored UAP generation for the DRL-based IDS. In addition, this is the first work that focuses on developing a UAP against a DRL-based IDS under realistic domain constraints based on not only the basic domain rules but also mathematical relations between the features. Furthermore, we enhance the evasion performance of the proposed UAP, by introducing a customized loss function based on the Pearson Correlation Coefficient, and we denote it as Customized UAP. To the best of our knowledge, this is also the first work using the PCC value in the UAP generation, even in the broader context. Four additional established UAP baselines are implemented for a comprehensive comparison. Experimental results demonstrate that our proposed Customized UAP outperforms two input-dependent attacks including Fast Gradient Sign Method (FGSM), Basic Iterative Method (BIM), and four UAP baselines, highlighting its effectiveness for real-world adversarial scenarios.

This paper addresses a critical gap in Multi-Objective Multi-Agent Reinforcement Learning (MOMARL) by introducing the first dedicated inner-loop actor-critic framework for continuous state and action spaces: Multi-Objective Multi-Agent Actor-Critic (MOMA-AC). Building on single-objective, single-agent algorithms, we instantiate this framework with Twin Delayed Deep Deterministic Policy Gradient (TD3) and Deep Deterministic Policy Gradient (DDPG), yielding MOMA-TD3 and MOMA-DDPG. The framework combines a multi-headed actor network, a centralised critic, and an objective preference-conditioning architecture, enabling a single neural network to encode the Pareto front of optimal trade-off policies for all agents across conflicting objectives in a continuous MOMARL setting. We also outline a natural test suite for continuous MOMARL by combining a pre-existing multi-agent single-objective physics simulator with its multi-objective single-agent counterpart. Evaluating cooperative locomotion tasks in this suite, we show that our framework achieves statistically significant improvements in expected utility and hypervolume relative to outer-loop and independent training baselines, while demonstrating stable scalability as the number of agents increases. These results establish our framework as a foundational step towards robust, scalable multi-objective policy learning in continuous multi-agent domains.

We present ContagionRL, a Gymnasium-compatible reinforcement learning platform specifically designed for systematic reward engineering in spatial epidemic simulations. Unlike traditional agent-based models that rely on fixed behavioral rules, our platform enables rigorous evaluation of how reward function design affects learned survival strategies across diverse epidemic scenarios. ContagionRL integrates a spatial SIRS+D epidemiological model with configurable environmental parameters, allowing researchers to stress-test reward functions under varying conditions including limited observability, different movement patterns, and heterogeneous population dynamics. We evaluate five distinct reward designs, ranging from sparse survival bonuses to a novel potential field approach, across multiple RL algorithms (PPO, SAC, A2C). Through systematic ablation studies, we identify that directional guidance and explicit adherence incentives are critical components for robust policy learning. Our comprehensive evaluation across varying infection rates, grid sizes, visibility constraints, and movement patterns reveals that reward function choice dramatically impacts agent behavior and survival outcomes. Agents trained with our potential field reward consistently achieve superior performance, learning maximal adherence to non-pharmaceutical interventions while developing sophisticated spatial avoidance strategies. The platform's modular design enables systematic exploration of reward-behavior relationships, addressing a knowledge gap in models of this type where reward engineering has received limited attention. ContagionRL is an effective platform for studying adaptive behavioral responses in epidemic contexts and highlight the importance of reward design, information structure, and environmental predictability in learning.

Carrier aggregation (CA) is a technique that allows mobile networks to combine multiple carriers to increase user data rate. On the uplink, for power constrained users, this translates to the need for an efficient resource allocation scheme, where each user distributes its available power among its assigned uplink carriers. Choosing a good set of carriers and allocating appropriate power on the carriers is important. If the carrier allocation on the uplink is such that a harmonic of a user's uplink carrier falls on the downlink frequency of that user, it leads to a self coupling-induced sensitivity degradation of that user's downlink receiver. In this paper, we model the uplink carrier aggregation problem as an optimal resource allocation problem with the associated constraints of non-linearities induced self interference (SI). This involves optimization over a discrete variable (which carriers need to be turned on) and a continuous variable (what power needs to be allocated on the selected carriers) in dynamic environments, a problem which is hard to solve using traditional methods owing to the mixed nature of the optimization variables and the additional need to consider the SI constraint. We adopt a reinforcement learning (RL) framework involving a compound-action actor-critic (CA2C) algorithm for the uplink carrier aggregation problem. We propose a novel reward function that is critical for enabling the proposed CA2C algorithm to efficiently handle SI. The CA2C algorithm along with the proposed reward function learns to assign and activate suitable carriers in an online fashion. Numerical results demonstrate that the proposed RL based scheme is able to achieve higher sum throughputs compared to naive schemes. The results also demonstrate that the proposed reward function allows the CA2C algorithm to adapt the optimization both in the presence and absence of SI.

Face anti-spoofing (FAS) has recently advanced in multimodal fusion, cross-domain generalization, and interpretability. With large language models and reinforcement learning (RL), strategy-based training offers new opportunities to jointly model these aspects. However, multimodal reasoning is more complex than unimodal reasoning, requiring accurate feature representation and cross-modal verification while facing scarce, high-quality annotations, which makes direct application of RL sub-optimal. We identify two key limitations of supervised fine-tuning plus RL (SFT+RL) for multimodal FAS: (1) limited multimodal reasoning paths restrict the use of complementary modalities and shrink the exploration space after SFT, weakening the effect of RL; and (2) mismatched single-task supervision versus diverse reasoning paths causes reasoning confusion, where models may exploit shortcuts by mapping images directly to answers and ignoring the intended reasoning. To address this, we propose PA-FAS, which enhances reasoning paths by constructing high-quality extended reasoning sequences from limited annotations, enriching paths and relaxing exploration constraints. We further introduce an answer-shuffling mechanism during SFT to force comprehensive multimodal analysis instead of using superficial cues, thereby encouraging deeper reasoning and mitigating shortcut learning. PA-FAS significantly improves multimodal reasoning accuracy and cross-domain generalization, and better unifies multimodal fusion, generalization, and interpretability for trustworthy FAS.

Associative thinking--the ability to connect seemingly unrelated ideas--is a foundational element of human creativity and problem-solving. This paper explores whether reinforcement learning (RL) guided by associative thinking principles can enhance a model's performance across diverse generative tasks, including story writing, code generation, and chart creation. We introduce a reinforcement learning framework that uses a prompt-based evaluation mechanism, incorporating established divergent thinking metrics from creativity research. A base language model is fine-tuned using this framework to reward outputs demonstrating higher novelty through higher degrees of conceptual connectivity. Interestingly, the experimental results suggest that RL-based associative thinking-trained models not only generate more original and coherent stories but also exhibit improved abstraction and flexibility in tasks such as programming and data visualization. Our findings provide initial evidence that modeling cognitive creativity principles through reinforcement learning can yield more adaptive and generative AI.

The incorporation of online reinforcement learning (RL) into diffusion and flow-based generative models has recently gained attention as a powerful paradigm for aligning model behavior with human preferences. By leveraging stochastic sampling via Stochastic Differential Equations (SDEs) during the denoising phase, these models can explore a variety of denoising trajectories, enhancing the exploratory capacity of RL. However, despite their ability to discover potentially high-reward samples, current approaches often struggle to effectively align with preferences due to the sparsity and narrowness of reward feedback. To overcome this limitation, we introduce a novel framework called Granular-GRPO (G$^2$RPO), which enables fine-grained and comprehensive evaluation of sampling directions in the RL training of flow models. Specifically, we propose a Singular Stochastic Sampling mechanism that supports step-wise stochastic exploration while ensuring strong correlation between injected noise and reward signals, enabling more accurate credit assignment to each SDE perturbation. Additionally, to mitigate the bias introduced by fixed-granularity denoising, we design a Multi-Granularity Advantage Integration module that aggregates advantages computed across multiple diffusion scales, resulting in a more robust and holistic assessment of sampling trajectories. Extensive experiments on various reward models, including both in-domain and out-of-domain settings, demonstrate that our G$^2$RPO outperforms existing flow-based GRPO baselines, highlighting its effectiveness and generalization capability.

Advanced reasoning in LLMs on challenging domains like mathematical reasoning can be tackled using verifiable rewards based reinforced fine-tuning (ReFT). In standard ReFT frameworks, a behavior model generates multiple completions with answers per problem, for the answer to be then scored by a reward function. While such RL post-training methods demonstrate significant performance improvements across challenging reasoning domains, the computational cost of generating completions during training with multiple inference steps makes the training cost non-trivial. To address this, we draw inspiration from off-policy RL, and speculative decoding to introduce a novel ReFT framework, dubbed Nested-ReFT, where a subset of layers of the target model acts as the behavior model to generate off-policy completions during training. The behavior model configured with dynamic layer skipping per batch during training decreases the inference cost compared to the standard ReFT frameworks. Our theoretical analysis shows that Nested-ReFT yields unbiased gradient estimates with controlled variance. Our empirical analysis demonstrates improved computational efficiency measured as tokens/sec across multiple math reasoning benchmarks and model sizes. Additionally, we explore three variants of bias mitigation to minimize the off-policyness in the gradient updates that allows for maintaining performance that matches the baseline ReFT performance.

User feedback is critical for refining recommendation systems, yet explicit feedback (e.g., likes or dislikes) remains scarce in practice. As a more feasible alternative, inferring user preferences from massive implicit feedback has shown great potential (e.g., a user quickly skipping a recommended video usually indicates disinterest). Unfortunately, implicit feedback is often noisy: a user might skip a video due to accidental clicks or other reasons, rather than disliking it. Such noise can easily misjudge user interests, thereby undermining recommendation performance. To address this issue, we propose a novel Group-aware User Behavior Simulation (G-UBS) paradigm, which leverages contextual guidance from relevant user groups, enabling robust and in-depth interpretation of implicit feedback for individual users. Specifically, G-UBS operates via two key agents. First, the User Group Manager (UGM) effectively clusters users to generate group profiles utilizing a ``summarize-cluster-reflect" workflow based on LLMs. Second, the User Feedback Modeler (UFM) employs an innovative group-aware reinforcement learning approach, where each user is guided by the associated group profiles during the reinforcement learning process, allowing UFM to robustly and deeply examine the reasons behind implicit feedback. To assess our G-UBS paradigm, we have constructed a Video Recommendation benchmark with Implicit Feedback (IF-VR). To the best of our knowledge, this is the first multi-modal benchmark for implicit feedback evaluation in video recommendation, encompassing 15k users, 25k videos, and 933k interaction records with implicit feedback. Extensive experiments on IF-VR demonstrate that G-UBS significantly outperforms mainstream LLMs and MLLMs, with a 4.0% higher proportion of videos achieving a play rate > 30% and 14.9% higher reasoning accuracy on IF-VR.