QMIX is a popular $Q$-learning algorithm for cooperative MARL in the centralised training and decentralised execution paradigm. In order to enable easy decentralisation, QMIX restricts the joint action $Q$-values it can represent to be a monotonic mixing of each agent's utilities. However, this restriction prevents it from representing value functions in which an agent's ordering over its actions can depend on other agents' actions. To analyse this representational limitation, we first formalise the objective QMIX optimises, which allows us to view QMIX as an operator that first computes the $Q$-learning targets and then projects them into the space representable by QMIX. This projection returns a representable $Q$-value that minimises the unweighted squared error across all joint actions. We show in particular that this projection can fail to recover the optimal policy even with access to $Q^*$, which primarily stems from the equal weighting placed on each joint action. We rectify this by introducing a weighting into the projection, in order to place more importance on the better joint actions. We propose two weighting schemes and prove that they recover the correct maximal action for any joint action $Q$-values, and therefore for $Q^*$ as well. Based on our analysis and results in the tabular setting we introduce two scalable versions of our algorithm, Centrally-Weighted (CW) QMIX and Optimistically-Weighted (OW) QMIX and demonstrate improved performance on both predator-prey and challenging multi-agent StarCraft benchmark tasks (Samvelyan et al., 2019).

The growing prevalence of artificial intelligence (AI) in various applications underscores the need for agents that can successfully navigate and adapt to an ever-changing, open-ended world. A key challenge is ensuring these AI agents are robust, excelling not only in familiar settings observed during training but also effectively generalising to previously unseen and varied scenarios. In this thesis, we harness methodologies from open-endedness and multi-agent learning to train and evaluate robust AI agents capable of generalising to novel environments, out-of-distribution inputs, and interactions with other co-player agents. We begin by introducing MiniHack, a sandbox framework for creating diverse environments through procedural content generation. Based on the game of NetHack, MiniHack enables the construction of new tasks for reinforcement learning (RL) agents with a focus on generalisation. We then present Maestro, a novel approach for generating adversarial curricula that progressively enhance the robustness and generality of RL agents in two-player zero-sum games. We further probe robustness in multi-agent domains, utilising quality-diversity methods to systematically identify vulnerabilities in state-of-the-art, pre-trained RL policies within the complex video game football domain, characterised by intertwined cooperative and competitive dynamics. Finally, we extend our exploration of robustness to the domain of LLMs. Here, our focus is on diagnosing and enhancing the robustness of LLMs against adversarial prompts, employing evolutionary search to generate a diverse range of effective inputs that aim to elicit undesirable outputs from an LLM. This work collectively paves the way for future advancements in AI robustness, enabling the development of agents that not only adapt to an ever-evolving world but also thrive in the face of unforeseen challenges and interactions.

We consider the problem of multiple agents sensing and acting in environments with the goal of maximising their shared utility. In these environments, agents must learn communication protocols in order to share information that is needed to solve the tasks. By embracing deep neural networks, we are able to demonstrate end-to-end learning of protocols in complex environments inspired by communication riddles and multi-agent computer vision problems with partial observability. We propose two approaches for learning in these domains: Reinforced Inter-Agent Learning (RIAL) and Differentiable Inter-Agent Learning (DIAL). The former uses deep Q-learning, while the latter exploits the fact that, during learning, agents can backpropagate error derivatives through (noisy) communication channels. Hence, this approach uses centralised learning but decentralised execution. Our experiments introduce new environments for studying the learning of communication protocols and present a set of engineering innovations that are essential for success in these domains.

The proof of your theory lacks discussion of POMDP settings. Although the framework in focused in solving the Dec-POMDP problem, most parts of the proof are under MDP setting. But there is no more discussion on that phenomenon. The use of weighting is not that convinced. In Section 6.2.3, the performance of the Weighted QMIX method is unacceptable.

I want to thank the authors for preparing the detailed rebuttal. This paper was discussed among all the reviewers during the post-rebuttal discussion phase. Also, given the borderline scores, we requested an additional emergency reviewer for this paper. While the rebuttal helped clarify some of the reviewers' questions, the reviewers shared a few concerns regarding the experimental evaluation, comparisons to SOTA, and the relationship of the proposed approach w.r.t. the relevant literature. Overall, the reviewers have a positive assessment of the paper and appreciated the technical insights to design the weighted QMIX algorithm.

Though the paper contains a very thorough experimental evaluation of the suggested DIAL technique for multi-agent settings and the reviewer understands that it would have taken a lot of time and effort to set up and evaluate the experiments, the paper does not make a very novel contribution. It is clear that idea of shared memory and passing message gradients between agents would speed up learning and help to find the optimal policy faster, but this might not be a very natural way to do it. For instance, humans working in teams do not have shared memories. Also, for humans, messages from other humans are a part of their observation at each time step, rather than separate signals which are treated specially as messages and optimized differently than the rest of the observation. The idea of passing message gradients is certainly useful to have trainable message protocols while training a set of agents to perform a repetitive task, but doesn't offer much insight or useful interpretation as to how humans perform tasks in teams.

QMIX is a popular Q -learning algorithm for cooperative MARL in the centralised training and decentralised execution paradigm. In order to enable easy decentralisation, QMIX restricts the joint action Q -values it can represent to be a monotonic mixing of each agent's utilities. However, this restriction prevents it from representing value functions in which an agent's ordering over its actions can depend on other agents' actions. To analyse this representational limitation, we first formalise the objective QMIX optimises, which allows us to view QMIX as an operator that first computes the Q -learning targets and then projects them into the space representable by QMIX. This projection returns a representable Q -value that minimises the unweighted squared error across all joint actions.

Establishing sound experimental standards and rigour is important in any growing field of research. Deep Multi-Agent Reinforcement Learning (MARL) is one such nascent field. Although exciting progress has been made, MARL has recently come under scrutiny for replicability issues and a lack of standardised evaluation methodology, specifically in the cooperative setting. Although protocols have been proposed to help alleviate the issue, it remains important to actively monitor the health of the field. In this work, we extend the database of evaluation methodology previously published by containing meta-data on MARL publications from top-rated conferences and compare the findings extracted from this updated database to the trends identified in their work. Our analysis shows that many of the worrying trends in performance reporting remain. This includes the omission of uncertainty quantification, not reporting all relevant evaluation details and a narrowing of algorithmic development classes. Promisingly, we do observe a trend towards more difficult scenarios in SMAC-v1, which if continued into SMAC-v2 will encourage novel algorithmic development. Our data indicate that replicability needs to be approached more proactively by the MARL community to ensure trust in the field as we move towards exciting new frontiers.

Responsible AI has risen to the forefront of the AI research community. As neural network-based learning algorithms continue to permeate real-world applications, the field of Responsible AI has played a large role in ensuring that such systems maintain a high-level of human-compatibility. Despite this progress, the state of the art in Responsible AI has ignored one crucial point: human problems are multi-agent problems. Predominant approaches largely consider the performance of a single AI system in isolation, but human problems are, by their very nature, multi-agent. From driving in traffic to negotiating economic policy, human problem-solving involves interaction and the interplay of the actions and motives of multiple individuals. This dissertation develops the study of responsible emergent multi-agent behavior, illustrating how researchers and practitioners can better understand and shape multi-agent learning with respect to three pillars of Responsible AI: interpretability, fairness, and robustness. First, I investigate multi-agent interpretability, presenting novel techniques for understanding emergent multi-agent behavior at multiple levels of granularity. With respect to low-level interpretability, I examine the extent to which implicit communication emerges as an aid to coordination in multi-agent populations. I introduce a novel curriculum-driven method for learning high-performing policies in difficult, sparse reward environments and show through a measure of position-based social influence that multi-agent teams that learn sophisticated coordination strategies exchange significantly more information through implicit signals than lesser-coordinated agents. Then, at a high-level, I study concept-based interpretability in the context of multi-agent learning. I propose a novel method for learning intrinsically interpretable, concept-based policies and show that it enables...

The research of extending deep reinforcement learning (drl) to multi-agent field has solved many complicated problems and made great achievements. However, almost all these studies only focus on discrete or continuous action space and there are few works having ever used multi-agent deep reinforcement learning to real-world environment problems which mostly have a hybrid action space. Therefore, in this paper, we propose two algorithms: deep multi-agent hybrid soft actor-critic (MAHSAC) and multi-agent hybrid deep deterministic policy gradients (MAHDDPG) to fill this gap. This two algorithms follow the centralized training and decentralized execution (CTDE) paradigm and could handle hybrid action space problems. Our experiences are running on multi-agent particle environment which is an easy multi-agent particle world, along with some basic simulated physics. The experimental results show that these algorithms have good performances.