Maximal Clique Based Distributed Coalition Formation for Task Allocation in Large-Scale Multi-agent Systems

Distributed task allocation in social networks

“…Their focus is on modeling the decision process of just a single mediator. Another approach is to partition the network into cliques of nodes, representing coalitions which the agents involved may use as a coordination mechanism [20]. The focus of that work is distributed coalition formation among agents, but in our approach, we do not need agents to form groups before allocating tasks.…”

Graph Patterns, Reinforcement Learning and Models of Reputation for Improving Coalition Formation in Collaborative Multi-agent Systems

“…Local communication constraints and desire to achieve the highest level of system-level robustness and faulttolerance have motivated the (maximal) clique based solutions, in which the resulting coalitions are required to be (maximal) cliques of the graph capturing the underlying ad hoc network topology of interacting agents. One such fully distributed and local graph partitioning algorithm for coalition formation, MCDCF, was originally proposed by us back in 2004 [1,2]. That algorithm has been demonstrated to be scalable and efficient on fairly large networks of cooperating agents (made of hundreds or even thousands of nodes), as long as those networks are sparse [1,2,7,8].…”

“…However, even in such complex, non-episodic environments, an agent should be able to learn over time, based on the feedback (typically, in the form of received utility or payoff) from the previous interactions. In our coalition formation for distributed task and/or resource allocation setting, this payoff comes from the utility associated with a completed task (or multiple tasks), that an agent was capable of completing as a member of a coalition it has joined [2,5]. That is, how successful an agent has been in striving to join best possible coalitions, is measured by the value (and implicitly, the success rate) of tasks that have been completed by the coalitions that this agent was a part of.…”

“…One such fully distributed and local graph partitioning algorithm for coalition formation, MCDCF, was originally proposed by us back in 2004 [1,2]. That algorithm has been demonstrated to be scalable and efficient on fairly large networks of cooperating agents (made of hundreds or even thousands of nodes), as long as those networks are sparse [1,2,7,8]. Further, the MCDCF algorithm, originally inspired by coordinating micro-UAVs [4], has a number of desirable properties relevant to a much broader range of collaborative MAS domains.…”

“…The proposed research direction builds on our prior work on distributed coalition formation in collaborative multiagent systems (MAS), centered at partitioning the underlying network of agents into non-overlapping cliques. At the core of that prior research is the MCDCF algorithm which provides a semantically simple, fully decentralized, local and (for sufficiently sparse networks) scalable mechanism for multi-agent coalition formation [1][2][3]. Our goal is to extend the MCDCF-based coalition formation along several new dimensions.…”