Multi-Agent Deep Reinforcement Learning (MADRL) was proven efficient in solving complex problems in robotics or games, yet most of the trained models are hard to interpret. While learning intrinsically interpretable models remains a prominent approach, its scalability and flexibility are limited in handling complex tasks or multi-agent dynamics. This paper advocates for direct interpretability, generating post hoc explanations directly from trained models, as a versatile and scalable alternative, offering insights into agents' behaviour, emergent phenomena, and biases without altering models' architectures. We explore modern methods, including relevance backpropagation, knowledge edition, model steering, activation patching, sparse autoencoders and circuit discovery, to highlight their applicability to single-agent, multi-agent, and training process challenges. By addressing MADRL interpretability, we propose directions aiming to advance active topics such as team identification, swarm coordination and sample efficiency.

Federated Reinforcement Learning (FedRL) improves sample efficiency Despite its promise, most FedRL frameworks [8, 10, 18, 50] operate while preserving privacy; however, most existing studies under the assumption of agent homogeneity (i.e., identical assume homogeneous agents, limiting its applicability in real-world policy networks and training configurations), which significantly scenarios. This paper investigates FedRL in black-box settings with limits FedRL's applicability in real-world scenarios. This limitation heterogeneous agents, where each agent employs distinct policy is particularly acute in resource-constrained environments, such as networks and training configurations without disclosing their internal in edge environments, where agents have limited power and need details. Knowledge Distillation (KD) is a promising method to adapt network structures and training strategies based on their for facilitating knowledge sharing among heterogeneous models, operational conditions to achieve effective training [47]. In addition, but it faces challenges related to the scarcity of public datasets and existing FedRL frameworks typically operate under a white-box limitations in knowledge representation when applied to FedRL. To paradigm, where models are openly shared among participants.

Overestimation arising from selecting unseen actions during policy evaluation is a major challenge in offline reinforcement learning (RL). A minimalist approach in the single-agent setting -- adding behavior cloning (BC) regularization to existing online RL algorithms -- has been shown to be effective; however, this approach is understudied in multi-agent settings. In particular, overestimation becomes worse in multi-agent settings due to the presence of multiple actions, resulting in the BC regularization-based approach easily suffering from either over-regularization or critic divergence. To address this, we propose a simple yet effective method, Behavior Cloning regularization with Critic Clipping (B3C), which clips the target critic value in policy evaluation based on the maximum return in the dataset and pushes the limit of the weight on the RL objective over BC regularization, thereby improving performance. Additionally, we leverage existing value factorization techniques, particularly non-linear factorization, which is understudied in offline settings. Integrated with non-linear value factorization, B3C outperforms state-of-the-art algorithms on various offline multi-agent benchmarks.

Detecting biases in the outputs produced by generative models is essential to reduce the potential risks associated with their application in critical settings. However, the majority of existing methodologies for identifying biases in generated text consider the models in isolation and neglect their contextual applications. Specifically, the biases that may arise in multi-agent systems involving generative models remain under-researched. To address this gap, we present a framework designed to quantify biases within multi-agent systems of conversational Large Language Models (LLMs). Our approach involves simulating small echo chambers, where pairs of LLMs, initialized with aligned perspectives on a polarizing topic, engage in discussions. Contrary to expectations, we observe significant shifts in the stance expressed in the generated messages, particularly within echo chambers where all agents initially express conservative viewpoints, in line with the well-documented political bias of many LLMs toward liberal positions. Crucially, the bias observed in the echo-chamber experiment remains undetected by current state-of-the-art bias detection methods that rely on questionnaires. This highlights a critical need for the development of a more sophisticated toolkit for bias detection and mitigation for AI multi-agent systems. The code to perform the experiments is publicly available at https://anonymous.4open.science/r/LLMsConversationalBias-7725.

Learning-to-Defer (L2D) facilitates optimal task allocation between AI systems and decision-makers. Despite its potential, we show that current two-stage L2D frameworks are highly vulnerable to adversarial attacks, which can misdirect queries or overwhelm decision agents, significantly degrading system performance. This paper conducts the first comprehensive analysis of adversarial robustness in two-stage L2D frameworks. We introduce two novel attack strategies -- untargeted and targeted -- that exploit inherent structural vulnerabilities in these systems. To mitigate these threats, we propose SARD, a robust, convex, deferral algorithm rooted in Bayes and $(\mathcal{R},\mathcal{G})$-consistency. Our approach guarantees optimal task allocation under adversarial perturbations for all surrogates in the cross-entropy family. Extensive experiments on classification, regression, and multi-task benchmarks validate the robustness of SARD.

Large Language Models (LLMs) can perform chart question-answering tasks but often generate unverified hallucinated responses. Existing answer attribution methods struggle to ground responses in source charts due to limited visual-semantic context, complex visual-text alignment requirements, and difficulties in bounding box prediction across complex layouts. We present ChartCitor, a multi-agent framework that provides fine-grained bounding box citations by identifying supporting evidence within chart images. The system orchestrates LLM agents to perform chart-to-table extraction, answer reformulation, table augmentation, evidence retrieval through pre-filtering and re-ranking, and table-to-chart mapping. ChartCitor outperforms existing baselines across different chart types. Qualitative user studies show that ChartCitor helps increase user trust in Generative AI by providing enhanced explainability for LLM-assisted chart QA and enables professionals to be more productive.

Monte Carlo Tree Search (MCTS) has proven highly effective in solving complex planning tasks by balancing exploration and exploitation using Upper Confidence Bound for Trees (UCT). However, existing work have not considered MCTS-based lifelong planning, where an agent faces a non-stationary series of tasks -- e.g., with varying transition probabilities and rewards -- that are drawn sequentially throughout the operational lifetime. This paper presents LiZero for Lipschitz lifelong planning using MCTS. We propose a novel concept of adaptive UCT (aUCT) to transfer knowledge from a source task to the exploration/exploitation of a new task, depending on both the Lipschitz continuity between tasks and the confidence of knowledge in in Monte Carlo action sampling. We analyze LiZero's acceleration factor in terms of improved sampling efficiency and also develop efficient algorithms to compute aUCT in an online fashion by both data-driven and model-based approaches, whose sampling complexity and error bounds are also characterized. Experiment results show that LiZero significantly outperforms existing MCTS and lifelong learning baselines in terms of much faster convergence (3$\sim$4x) to optimal rewards. Our results highlight the potential of LiZero to advance decision-making and planning in dynamic real-world environments.

The increasing availability of computer-mediated knowledge and the advancement of information and communication technologies have altered the methods through which health care information is sought [3] [25] [30]. The Internet has had a significant impact on healthcare service and is a virtual medical library for an estimated 75-80% of users in developed countries [4] [5] [11]. On an average day, more than six million patients and their caregivers in the United States use the Internet to obtain health and medical information. This number exceeds the average daily number of 2.27 million Americans who make visits to physician offices [11] [18] [26]. Furthermore, not only patients but their caregivers want to get actively involved in the health-care management of their loved ones. In a research nearly 60% of people who identified themselves as caregivers use the Internet to find answers to their health-related questions [16]. This computer mediated environment has become, as Vargo and Lusch [32] argue, a fundamental hub where "people exchange to acquire the benefits of specialized competencies (knowledge and skills), or services."

The significant role of division of labor (DOL) in promoting cooperation is widely recognized in real-world applications.Many cooperative multi-agent reinforcement learning (MARL) methods have incorporated the concept of DOL to improve cooperation among agents.However, the tasks used in existing testbeds typically correspond to tasks where DOL is often not a necessary feature for achieving optimal policies.Additionally, the full utilize of DOL concept in MARL methods remains unrealized due to the absence of appropriate tasks.To enhance the generality and applicability of MARL methods in real-world scenarios, there is a necessary to develop tasks that demand multi-agent DOL and cooperation.In this paper, we propose a series of tasks designed to meet these requirements, drawing on real-world rules as the guidance for their design.We guarantee that DOL and cooperation are necessary condition for completing tasks and introduce three factors to expand the diversity of proposed tasks to cover more realistic situations.We evaluate 10 cooperative MARL methods on the proposed tasks.The results indicate that all baselines perform poorly on these tasks.To further validate the solvability of these tasks, we also propose simplified variants of proposed tasks.Experimental results show that baselines are able to handle these simplified variants, providing evidence of the solvability of the proposed tasks.The source files is available at https://github.com/Yurui-Li/CTC.

We study the uniform-in-time weak propagation of chaos for the consensus-based optimization (CBO) method on a bounded searching domain. We apply the methodology for studying long-time behaviors of interacting particle systems developed in the work of Delarue and Tse (ArXiv:2104.14973). Our work shows that the weak error has order $O(N^{-1})$ uniformly in time, where $N$ denotes the number of particles. The main strategy behind the proofs are the decomposition of the weak errors using the linearized Fokker-Planck equations and the exponential decay of their Sobolev norms. Consequently, our result leads to the joint convergence of the empirical distribution of the CBO particle system to the Dirac-delta distribution at the global minimizer in population size and running time in Wasserstein-type metrics.