Search

This paper develops a new approach to speed up Generalized Distributive Law (GDL) based message passing algorithms that are used to solve large-scale Distributed Constraint Optimization Problems (DCOPs) in multi-agent systems. In particular, we significantly reduce computation and communication costs in terms of convergence time for algorithms such as Max-Sum, Bounded Max-Sum, Fast Max-Sum, Bounded Fast Max-Sum, BnB Max-Sum, BnB Fast Max-Sum and Generalized Fast Belief Propagation. This is important since it is often observed that the outcome obtained from such algorithms becomes outdated or unusable if the optimization process takes too much time. Specifically, the issue of taking too long to complete the internal operation of a DCOP algorithm is even more severe and commonplace in a system where the algorithm has to deal with a large number of agents, tasks and resources. This, in turn, limits the practical scalability of such algorithms. In other words, an optimization algorithm can be used in larger systems if the completion time can be reduced. However, it is challenging to maintain the solution quality while minimizing the completion time. Considering this trade-off, we propose a generic message passing protocol for GDL-based algorithms that combines clustering with domain pruning, as well as the use of a regression method to determine the appropriate number of clusters for a given scenario. We empirically evaluate the performance of our method in a number of settings and find that it brings down the completion time by around 37–85% (1.6–6.5 times faster) for 100–900 nodes, and by around 47–91% (1.9–11 times faster) for 3000–10 000 nodes compared to the current state-of-the-art. [ABSTRACT FROM AUTHOR]

Research on optimization in multi-agent systems (MASs) has contributed with a wealth of techniques to solve many of the challenges arising in a wide range of multi-agent application domains. Multi-agent optimization focuses on casting MAS problems into optimization problems. The solving of those problems could possibly involve the active participation of the agents in a MAS. Research on multi-agent optimization has rapidly become a very technical, specialized field. Moreover, the contributions to the field in the literature are largely scattered. These two factors dramatically hinder access to a basic, general view of the foundations of the field. This tutorial is intended to ease such access by providing a gentle introduction to fundamental concepts and techniques on multi-agent optimization. [ABSTRACT FROM AUTHOR]

This paper tackles the problem of parallelizing heterogeneous computational tasks across a number of computational nodes (aka agents) where each agent may not be able to perform all the tasks and may have different computational speeds. An equivalent problem can be found in operations research, and it is known as scheduling tasks on unrelated parallel machines (also known as R∥Cmax). Given this equivalence observation, we present the spanning tree decentralized task distribution algorithm (ST-DTDA), the first decentralized solution to R∥Cmax. ST-DTDA achieves decomposition by means of the min–max algorithm, a member of the generalized distributive law family, that performs inference by message-passing along the edges of a graphical model (known as a junction tree). Specifically, ST-DTDA uses min–max to optimally solve an approximation of the original R∥Cmax problem that results from eliminating possible agent-task allocations until it is mapped into an acyclic structure. To eliminate those allocations that are least likely to have an impact on the solution quality, ST-DTDA uses a heuristic approach. Moreover, ST-DTDA provides a per-instance approximation ratio that guarantees that the makespan of its solution (optimal in the approximated R∥Cmax problem) is not more than a factor ρ times the makespan of the optimal of the original problem. In our empirical evaluation of ST-DTDA, we show that ST-DTDA, with a min-regret heuristic, converges to solutions that are between 78 and 95% optimal whilst providing approximation ratios lower than 3. [ABSTRACT FROM AUTHOR]

Emergency responders are faced with a number of significant challenges when managing major disasters. First, the number of rescue tasks posed is usually larger than the number of responders (or agents) and the resources available to them. Second, each task is likely to require a different level of effort in order to be completed by its deadline. Third, new tasks may continually appear or disappear from the environment, thus requiring the responders to quickly recompute their allocation of resources. Fourth, forming teams or coalitions of multiple agents from different agencies is vital since no single agency will have all the resources needed to save victims, unblock roads and extinguish the fires which might erupt in the disaster space. Given this, coalitions have to be efficiently selected and scheduled to work across the disaster space so as to maximize the number of lives and the portion of the infrastructure saved. In particular, it is important that the selection of such coalitions should be performed in a decentralized fashion in order to avoid a single point of failure in the system. Moreover, it is critical that responders communicate only locally given they are likely to have limited battery power or minimal access to long-range communication devices. Against this background, we provide a novel decentralized solution to the coalition formation process that pervades disaster management. More specifically, we model the emergency management scenario defined in the RoboCup Rescue disaster simulation platform as a coalition formation with spatial and temporal constraints (CFST) problem where agents form coalitions to complete tasks, each with different demands. To design a decentralized algorithm for CFST, we formulate it as a distributed constraint optimization problem and show how to solve it using the state-of-the-art Max-Sum algorithm that provides a completely decentralized message-passing solution. We then provide a novel algorithm (F-Max-Sum) that avoids sending redundant messages and efficiently adapts to changes in the environment. In empirical evaluations, our algorithm is shown to generate better solutions than other decentralized algorithms used for this problem. [ABSTRACT FROM PUBLISHER]