Principal Investigators:
Lik Mui - Health Sciences & Technology
Prof. Peter Szolovits - Electrical Engineering & Computer Science (advisor)

Additional Investigators:
Dr. Mojdeh Mohtashemi, Waikit Koh, Michael McGeachie (EECS), Raj Krishnan (EECS), Francisco Tanudjaja, Cheewee Ang

Project Overview:

iMatch is a research project studying resource discovery in highly decentralized systems.

Users equipped with iMatch enabled PocketPCs will be able to dynamically locate resources corresponding to a match request. Underlying iMatch is a highly decentralized multi-agents framework (called iAgent). Each "autonomous system" for iMatch is an iAgent. Agents exist as applications on PDA's, desktops, or servers. When a user logs on and is authenticated locally by an iMatch agent, interaction between the local agent and the rest of iMatch occurs via KQML style messages. Each agent communicates with every other agent using one of two methods:

1. send messages using sockets (.NET remoting)
2. send messages using web services interfaces (HTTP/SOAP)

Since knowledge about each user is guarded by a local agent, revelation of user's profile or data is granted by owners in one of two ways:

1. local information can be set to one of three security levels:
   • public (available to all other agents)
   • friends (available to specified friend agents)
   • private (available only for local agent)
2. other agents can be classified in three categories
   • acquaintances (given "friends" access)
   • known-agents (referred agents from friends)
   • unknown agents (never heard of agents)

Certain central agents act as bootstrap agents to help new user agents start building a trust network of acquaintances. Trust for other agents is inferred based on a reputation framework based on Mui, et al, (2001). Once trust is established, matching user request to appropriate resources is performed based on a preference-based framework based on the Ceteris Paribus representation (c.f., Doyle, et al., 1994 and McGeachie, 2001).

We hope to build on this research to equip students staff members in academic environments with personal software agents. Each iMatch agent would help manage its owner's academic life through both static and dynamic profile matching, aiming to encourage collaboration. This collaboration can have several goals: completing final projects; studying for exams, or tutoring one another.

J. Doyle, M. P. Wellman (1994) "Representing Preferences as Ceteris Paribus Comparatives," Working Notes of the AAAI Symposium on Decision-Theoretic Planning.

M. McGeachie (2001) "Ceteris Paribus Preference Specification," LCS unpublished working paper.

L. Mui, M. Mohtashemi, C. Ang, P. Szolovits, A. Halberstadt (2001) "Bayesian Ratings in Distributed Systems: Theories, Models, and Simulations," MIT LCS Memorandum.

For More Information:

• iMatch web site