CUNY-GC Yuqing Tang's Homepage

Research

My research area is autonomous agents and multiagent systems, a sub-field of artificial intelligence. An agent is a computer system (or a physical entity controlled by a computer system, such as a robot) that is capable of acting independently on behalf of its user or owner. In other words, given its preset design objectives, an agent can decide what actions to take and execute them without being told explicitly what to do at any given moment. A multiagent system (MAS) is one that consists of a number of agents, which interact with one another, typically by exchanging messages through some computer network infrastructure. I am particularly working on a specific kind of interaction between agents in a multiagent system --- an argumentation-based dialogue mechanism. It is a communication mechanism between intelligent agents which is expected to be able to make use of incomplete and inconsistent information under decentralized control while promoting effective overall problem-solving in a coherent manner.

This research is motivated by the need for coherent problem solving in a decentralized multiagent system. As of today, more and more computers are being used to aid us in almost every aspect of our life and work. These computers usually behave independently but interact with one another. They can be in the form of a desktop or a laptop computer, in the form of a mobile phone or a personal digital assistant, in the form of a robot, vehicle that drives itself, or a space craft controlled by computers, and many other forms are being and will be produced. The functions of these computers range from helping us store and manage schedules, collecting information and generating reports for us from different sources, to controlling the life-critical devices (e.g. the space craft, nuclear plants, and autonomous driving vehicles). As computers are getting involved in human life more deeply and more broadly, the tasks they are handling for human are getting more and more diverse and complex while at the same time many of these tasks overlap with one another. To conquer the diversity and complexity, we usually employ the divide-and-conquer approach to decompose the systems into many sub-systems each of which is a computer or a computer controlled entity for each task. These computers are related to one another because of either the overlapping tasks they are handling, or the interrelation between their human owners. For this reason, interrelated computers are usually connected with one another through a computer network infrastructure. Arising from this decomposition approach and arising from the social nature of their human owners, the social aspects of these independent computers play a more and more important role in the system. Autonomous Agents and Multiagent Systems is a research area addressing the complexity and the social nature of these independent computer systems which can demonstrate overall organized problem-solving behaviors in a coherent manner.

In particular, I am working on a formal model of multiagent dialogue mechanisms. As the designers and the programmers of the agents are not omniscient and can make mistakes, and the perception sub-system of the agents are not perfect and not omniscient as well, incomplete information and inconsistent information are unavoidable. Non-deterministic state transitions and state-action tables (policies) are used to model the incomplete information about the external world, the behaviors of individual agents and the interactions of these agents. Argumentation-based reasoning is used to resolve inconsistent information within an agent as well as across agents.Specifically the argumentation approach is to resolve conflicts arising from agents’ different views of the environment dynamics, their different goals about what to achieve, and their different ways to achieve the goals. The approach is composed of two phases: argument construction and conflict pattern analysis. The argument construction is essentially an information clustering procedure in which information are gathered into arguments according to certain reasoning pattern (either a strict logic, a defeasible logic, statistical inferences, or their combinations). The conflict pattern analysis is a procedure to assign status (accepted, rejected, unknown, or other intermediate class of status) to the information clusters according to the conflict relation among these clusters with respect to the reasoning pattern being used. To address the complexity of designing and specifying dialogue protocols and conversation policies, I will borrow the idea of hierarchical task networks from AI planning to specify the decomposition of these protocols and conversation policies. The underlying non-deterministic state transitions, policies and argumentation processes will be specified by quantified boolean formulae (QBF) which are in turn implemented using binary decision diagrams (BDDs). The power of BDDs lies in its implicit and efficient symbolic manipulation of sets and relations. By using BDDs, I hope to be able to have an efficient implementation of this model. Several applications of this model will be tried out: multiagent public argumentation, multiagent planning, team plan execution and plan revision. Some intermediate results of this research can be found in my recent publications, and more results will be available shortly.

This research is not to solve all the problems at once, instead it is just to provide an initial attempt to layout the basic elements for one out of many possible dialogue mechanisms that are computationally simple enough for implementation, and flexible and powerful enough to carry out multiagent tasks with incomplete and inconsistent information under decentralized control.

Besides the dissertation research, I have been involved in several research projects in various areas (which were chosen to balance between my research interests and the fundings available at CUNY). On the whole, they are listed as follows:

Not limited to the above projects, all in all I am interested in: