|
|
Ping XuanResearch |
Assistant Professor Department of Mathematics and Computer Science Clark University 950 Main Street Worcester, MA 01610-1477 |
|
Description of My Main Research AreasToward a Richer Model of Agent Commitments:Since each agent only sees a partial view of the overall (multiagent) task network, they must rely on commitments established between the agents in order to formulate their plans and facilitate coordination. However, due to uncertain task outcomes, commitments may not always get fulfilled, which lead to bad performance. A quantified model of commitment, with richer semantics, is needed. Commitments themselves can be viewed as dynamic processes, and they encapsulate the information needed in the coordination part of multiagent planning.
Decentralized Multiagent Markov ProcessesDEC-MDPs and DEC-POMDPs are computationally very complex. This gives us strong incentives to search for approximation solutions and heuristics methods. While other researchers focus more on improving computational methods and classifying sub-classes of these complex problems, I believe we can also draw upon a lot of knowledge developed in traditional multiagent planning. The dichotomy of local reasoning and coordination will be the key for scalable models and solutions.
Communication and CoordinationOne way of handling uncertainty is to communicate, and sharing nonlocal information is central to coordination. Developing a more precise model of commitments is one aspect of this problem, but it is equally important to determine when to communicate. These communication decisions are tied to the agent's beliefs about agent commitments.
Issues in Agent Virtual OrganizationsManaging uncertainty is not limited at agent level - it is also essential at organizational level. I am interested at a variety of issues, such as plan robustness, organizational monitoring, and decommitments in multiagent organizations (collaboration with Professor S. Zhang's group at UMass Dartmouth.)
|
| pxuan@clarku.edu | 2005 |