Tsz-Chiu Au
Ph.D.
Department of Computer Science University of Maryland, College Park
I am currently a postdoc in the Department of Computer Science at the University of Texas at Austin. Previously I was a Ph.D. student in the Department of Computer Science at the University of Maryland. I received my B.Eng. degree in Computer Science from Hong Kong University of Science and Technology in Hong Kong. My supervisor is Prof. Peter Stone. This is my CV.

Research Interests: Multi-agent systems, artificial intelligent planning, case-based reasoning, computer game playing, and intelligent transportation system. Specifically, autonomous intersection management, error detection in multi-agent systems, opponent modeling, planning with volatile external information, planning under uncertainty, and Hierarchical Task Network (HTN) planning.

Current Project: Autonomous Intersection Management

Research Accomplishments: Coping with noise in non-zero-sum games. A noisy multiagent environment is one in which a “noise gremlin” can randomly change the agents' actions or communications. This causes errors in the interactions among agents, creating huge problems in many multiagent systems. For instance, in the famous Iterated Prisoner's Dilemma (IPD), even a small amount of noise can cause great difficulty for the agents. It is well-known that Tit-For-Tat, one of the best-known strategies in the IPD, performs badly in the presence of noise. To cope with this problem, I devised a technique called symbolic noise detection (SND) to detect and correct errors caused by noise. The idea behind SND is that when there are strong incentives to cooperate, agents often behave deterministically, and these deterministic behaviors can be used to detect noise. In the 20th Anniversary Iterated Prisoner's Dilemma competition, my SND agent placed third out of 165 agents in the “noise” category, and was the best performer among programs that had no “slave” programs feeding points to them. Synthesis of strategies from interaction traces. To create better agents in multiagent environments, one may want to examine the observed behaviors of existing agents in order to combine their best skills. But the agents being observed may exhibit incompatible behaviors---the agents choose to do different things in some situations---and it is not immediately clear which behavior is better. Even worse, when there are many agents, it is hard to tell which subset of the observed behaviors is the best and is free of incompatibilities. But once we identify the best subset we can combine the observed behaviors together to form a composite strategy that can potentially outperform all agents being observed. To examine this strategy synthesis scheme, I devised a technique for combining interaction traces that synthesizes new composite strategies by combining the best behaviors from records of interactions among many different agents. In cross-validated experiments involving more than 50 agents, composite strategies produced from these agents produced large improvements in the performance of nearly all of the agents. Managing volatile data for planning processes in semantic web service composition. An important issue in web service composition is information volatility---the collected information may change or expire during or after the composition process. The expiration of collected data is a problem because it can invalidate the results generated by web services. For example, a web-based trip planner can generate incorrect travel plans if it does not know the information provided by an airline company has changed. To address this problem, I proposed several query management strategies for handling volatile data for web services, and theoretically analyzed the conditions under which the solutions returned by the web service composition process will remain valid. My work is the first to look at this problem in the context of the semantic web.

Selected Publications: T.-C. Au, S. Kraus, and D. Nau. Synthesis of Strategies from Interaction Traces. Proceedings of the Seventh International Joint Conference on Autonomous Agents and Multiagent Systems (AAMAS'08), pp. 855-862, May 2008. [pdf] [data] [bibtex]. T.-C. Au and D. Nau. Is it Accidental or Intentional? A Symbolic Approach to the Noisy Iterated Prisoner's Dilemma. The Iterated Prisoners' Dilemma: 20 Years on, pp.231-262, World Scientific, 2007. [pdf] [bibtex]. T.-C. Au, S. Kraus, and D. Nau. Symbolic noise detection in the noisy iterated chicken game and the noisy iterated battle of the sexes. In First International Conference on Computational Cultural Dynamics (ICCCD-2007), August 2007. [pdf] [bibtex]. T.-C. Au. Dynamic Programming with Stochastic Opponent Models in Social Games: A Preliminary Report. In First International Conference on Computational Cultural Dynamics (ICCCD-2007), August 2007. [pdf] [bibtex]. T.-C. Au and D. Nau. Reactive Query Policies: A Formalism for Planning with Volatile External Information. IEEE Symposium on Computational Intelligence and Data Mining (CIDM), pp. 243-250, 2007. [pdf] [bibtex]. T.-C. Au and D. Nau. The Incompleteness of Planning with Volatile External Information. Proceedings of the European Conference on Artificial Intelligence (ECAI), August 2006. [pdf] [bibtex]. T.-C. Au, and D. Nau. Accident or Intention: That is the Question (in the Noisy Iterated Prisoner's Dilemma). Proceedings of the Fifth International Joint Conference on Autonomous Agents and Multiagent Systems (AAMAS'06). pp. 561-568, May 2006. [pdf] [bibtex]. (The corresponding NECTAR paper in AAAI-06: [pdf]) T.-C. Au, U. Kuter and D. Nau. Web Service Composition with Volatile Information. Proceedings of the 4th International Semantic Web Conference (ISWC-2005), pp. 52-66, 2005. [pdf] [bibtex]. T.-C. Au, D. Nau, and V. Subrahamanian. Utilizing volatile external information during planning. Proceedings of the European Conference on Artificial Intelligence (ECAI), pp. 647-651, August 2004. [pdf] [bibtex]. D. Nau, T.-C. Au, O. Ilghami, U. Kuter, H. Muņoz-Avila, J. W. Murdock, D. Wu, and F. Yaman. Applications of SHOP and SHOP2. IEEE Intelligent Systems 20:2, pp. 34-41, 2005. [pdf] [html] [bibtex]. D. Nau, T.-C. Au, O. Ilghami, U. Kuter, J. W. Murdock, D. Wu, and F. Yaman. SHOP2: An HTN planning system. Journal of Artificial Intelligence Research 20:379-404, December 2003. [pdf] [html] [bibtex]. H. Muņoz-Avila, D. S. Nau, and Tsz-Chiu Au. On the complexity of plan adaptation by derivational analogy in a universal classical planning framework. Proceedings of the European Conference on Case-Based Reasoning (ECCBR), pp. 13-27, September 4-7 2002. [pdf] [bibtex].

Education: Ph.D. in Department of Computer Science, University of Maryland, College Park M.S. in Department of Computer Science, University of Maryland, College Park B.Eng. in Computer Science, Hong Kong University of Science and Technology

Contact Information: Email: chiu (at) cs . utexas . edu Phone: (512) 232-4895 Address: Tsz-Chiu Au Department of Computer Science University of Texas at Austin 1 University Station Austin, TX 78712