Sparse rewards are one of the most important challenges in reinforcement learning. In the single-agent setting, these challenges have been addressed by introducing intrinsic rewards that motivate agents to explore unseen regions of their state spaces. Applying these techniques naively to the multi-agent setting results in individual agents exploring independently, without any coordination among themselves. We argue that learning in cooperative multi-agent settings can be accelerated and improved if agents coordinate with respect to what they have explored. In this paper we propose an approach for learning how to dynamically select between different types of intrinsic rewards which consider not just what an individual agent has explored, but all agents, such that the agents can coordinate their exploration and maximize extrinsic returns. Concretely, we formulate the approach as a hierarchical policy where a high-level controller selects among sets of policies trained on different types of intrinsic rewards and the low-level controllers learn the action policies of all agents under these specific rewards. We demonstrate the effectiveness of the proposed approach in a multi-agent learning domain with sparse rewards.

Reinforcement learning in multi-agent scenarios is important for real-world applications but presents challenges beyond those seen in single-agent settings. We present an actor-critic algorithm that trains decentralized policies in multi-agent settings, using centrally computed critics that share an attention mechanism which selects relevant information for each agent at every timestep. This attention mechanism enables more effective and scalable learning in complex multi-agent environments, when compared to recent approaches. Our approach is applicable not only to cooperative settings with shared rewards, but also individualized reward settings, including adversarial settings, as well as settings that do not provide global states, and it makes no assumptions about the action spaces of the agents. As such, it is flexible enough to be applied to most multi-agent learning problems.

We propose a novel, flexible algorithm for combining together metaheuristic optimizers for non-convex optimization problems. Our approach treats the constituent optimizers as a team of complex agents that communicate information amongst each other at various intervals during the simulation process. The information produced by each individual agent can be combined in various ways via higher-level operators. In our experiments on key benchmark functions, we investigate how the performance of our algorithm varies with respect to several of its key modifiable properties. Finally, we apply our proposed algorithm to classification problems involving the optimization of support-vector machine classifiers.

We develop model free PAC performance guarantees for multiple concurrent MDPs, extending recent works where a single learner interacts with multiple non-interacting agents in a noise free environment. Our framework allows noisy and resource limited communication between agents, and develops novel PAC guarantees in this extended setting. By allowing communication between the agents themselves, we suggest improved PAC-exploration algorithms that can overcome the communication noise and lead to improved sample complexity bounds. We provide a theoretically motivated algorithm that optimally combines information from the resource limited agents, thereby analyzing the interaction between noise and communication constraints that are ubiquitous in real-world systems. We present empirical results for a simple task that supports our theoretical formulations and improve upon naive information fusion methods.

Commitments capture how an agent relates to another agent, whereas goals describe states of the world that an agent is motivated to bring about. Commitments are elements of the social state of a set of agents whereas goals are elements of the private states of individual agents. It makes intuitive sense that goals and commitments are understood as being complementary to each other. More importantly, an agent's goals and commitments ought to be coherent, in the sense that an agent's goals would lead it to adopt or modify relevant commitments and an agent's commitments would lead it to adopt or modify relevant goals. However, despite the intuitive naturalness of the above connections, they have not been adequately studied in a formal framework. This article provides a combined operational semantics for goals and commitments by relating their respective life cycles as a basis for how these concepts (1) cohere for an individual agent and (2) engender cooperation among agents.

In this paper, we study the problem of forming coalitions with heterogeneous agents for allocating them to tasks. Several agents work together to complete a given task. Due to the inherent complexity of real-world tasks and limited capabilities of a particular type of a physical agent such as a robot, it is imperative to form a team consisting of different types of robots to complete the tasks. Our work in this paper proposes a distributed bipartite graph partitioning approach along with a region growing strategy for coalition formation with heterogeneous agents such as humans and/or robots for instantaneous allocation to tasks (ST-MR-IA). We also extend this approach to apply in the scenarios where the tasks might have dependencies among each other (ST-MR-TD).We have implemented the proposed algorithms within theWebots simulator. The proposed strategy allocates near-optimal (up to 98%) agent coalitions to tasks. Results also show that our proposed approach can easily handle as many as 100 agents and 10 tasks while spending an almost negligible amount of time.

During opinion formation, interacting agents can be assumed to be engaging in learning and decision-making processes to satisfy their individual goals. These goals are determined by the agents’ preferences – which are often unknown, complex and unpredictable. Most opinion formation frameworks however, assume static preferences and fail to model practical situations where human preferences change. We propose a new framework to simulate the process of opinion formation under uncertainty and dynamism. Agents who are unaware of their implicit con-textual preferences utilize inverse reinforcement learning to compute reward functions that determines their preferences. Reinforcement learning is subsequently used to optimize the agents’ behavior and satisfy their individual goals. The novelty of our approach lies in its ability to capture uncertainty and dynamism in the agent’s preferences, which are assumed to be unknown initially. This framework is compared to a baseline method based on reinforcement learning, and results show its ability to per-form better under dynamic scenarios.

We explore value-based solutions for multi-agent reinforcement learning (MARL) tasks in the centralized training with decentralized execution (CTDE) regime popularized recently. However, VDN and QMIX are representative examples that use the idea of factorization of the joint action-value function into individual ones for decentralized execution. VDN and QMIX address only a fraction of factorizable MARL tasks due to their structural constraint in factorization such as additivity and monotonicity. In this paper, we propose a new factorization method for MARL, QTRAN, which is free from such structural constraints and takes on a new approach to transforming the original joint action-value function into an easily factorizable one, with the same optimal actions. QTRAN guarantees more general factorization than VDN or QMIX, thus covering a much wider class of MARL tasks than does previous methods. Our experiments for the tasks of multi-domain Gaussian-squeeze and modified predator-prey demonstrate QTRAN's superior performance with especially larger margins in games whose payoffs penalize non-cooperative behavior more aggressively.

Self-adaptation has been proposed as a mechanism to counter complexity in control problems of technical systems. A major driver behind self-adaptation is the idea to transfer traditional design-time decisions to runtime and into the responsibility of systems themselves. In order to deal with unforeseen events and conditions, systems need creativity -- typically realized by means of machine learning capabilities. Such learning mechanisms are based on different sources of knowledge. Feedback from the environment used for reinforcement purposes is probably the most prominent one within the self-adapting and self-organizing (SASO) systems community. However, the impact of other (sub-)systems on the success of the individual system's learning performance has mostly been neglected in this context. In this article, we propose a novel methodology to identify effects of actions performed by other systems in a shared environment on the utility achievement of an autonomous system. Consider smart cameras (SC) as illustrating example: For goals such as 3D reconstruction of objects, the most promising configuration of one SC in terms of pan/tilt/zoom parameters depends largely on the configuration of other SCs in the vicinity. Since such mutual influences cannot be pre-defined for dynamic systems, they have to be learned at runtime. Furthermore, they have to be taken into consideration when self-improving the own configuration decisions based on a feedback loop concept, e.g., known from the SASO domain or the Autonomic and Organic Computing initiatives. We define a methodology to detect such influences at runtime, present an approach to consider this information in a reinforcement learning technique, and analyze the behavior in artificial as well as real-world SASO system settings.

An open problem in machine learning is whether flat minima generalize better and how to compute such minima efficiently. This is a very challenging problem. As a first step towards understanding this question we formalize it as an optimization problem with weakly interacting agents. We review appropriate background material from the theory of stochastic processes and provide insights that are relevant to practitioners. We propose an algorithmic framework for an extended stochastic gradient Langevin dynamics and illustrate its potential. The paper is written as a tutorial, and presents an alternative use of multi-agent learning. Our primary focus is on the design of algorithms for machine learning applications; however the underlying mathematical framework is suitable for the understanding of large scale systems of agent based models that are popular in the social sciences, economics and finance.