My general research interests include decision-theoretic artificial intelligence, temporal probabilistic reasoning, machine learning, and medical informatics. I am mostly interested in developing tools and techniques for facilitating medical decision making and medical problem solving . While such techniques may be generally applicable, medicine provides a rich and important domain for examining many complex and interesting issues.
In my research, I am designing and implementing advanced techniques for supporting dynamic decision making under uncertainty. My main task is to characterize the reasoning and the knowledge involved in dynamic decision modeling in medicine, which usually begins with an imprecise problem description, and ends with the solutions and insights provided by a formal stochastic model analysis. Besides developing and codifying an ontology of the modeling process, I am also designing a dynamic decision modeling language that generalizes existing frameworks. This research aims to provide physicians with a set of tools for analyzing difficult medical decisions. The resulting decision modeling ontology can potentially serve as guidelines for the expert physicians and as teaching aids for the novice physicians.
I am also interested in the management of complex and varied information in dynamic decision modeling. These include integrating AI and mathematical techniques for decision making in an unstructured domain, representing context-sensitive knowledge, and automated learning from multiple knowledge sources.
AI techniques are useful for formulating decision problems, i.e., determining the relevant decision factors, in domains that are difficult to be formally modeled. On the other hand, mathematical techniques in control theory and decision theory are effective for solving decision problems that can be formally formulated. In recent years, researchers in the field of knowledge-based decision systems have begun integrating the two types of techniques. My current research represents one of the initial efforts in this area that addresses the dynamic aspects of the decision problems. Some of the relevant issues include:
Incorporating domain-specific knowledge in a uniform manner for more effective automated formulation of decision models;
Developing some general methodologies, e.g., problem transformations or reformulations, that can relax the assumptions and limitations inherent in the mathematical models and solution methods;
Generalizing the dynamic decision modeling ontology to other problem types or domains, e.g., complex business decisions, government policy decisions, transportation planning, etc.; and
Identifying and resolving the practical bottlenecks that my system design will face when put into routine use.
Effective organization of symbolic and numerical knowledge is important for facilitating the integration of AI and mathematical techniques. I am interested in exploring how such knowledge can be represented in a context-sensitive manner. Context-sensitivity is essential for clear knowledge base semantics and efficient inferences; it is one of the most important and difficult tasks in knowledge representation research. By improving and adapting existing technologies, I hope to solve this problem in the specific domain of medical decision making. I will then explore if my solution can be generalized.
Medical decision making in practice makes use of a large amount of data. Such data are usually painstakingly extracted or interpreted from medical literature and patient records. The interpreted data are then combined with the physician's subjective judgment to produce the likelihoods or probabilities of various decision factors. Many hospitals are starting to provide large amount of computerized patient records (mostly anonymously) for research use. Given access to one or more of such data bases, I would like to devise some general techniques for extracting probabilities or temporal patterns from the data, e.g., calculating the time elapsed between a coronary bypass surgery to the onset of another episode of chest pain.
Other research areas that I am interested in are machine learning, distributed AI, and integrated hospital information systems. I am collaborating extensively with practicing physicians and colleagues with similar research interests.
I am very lucky to have some great mentors who have not only introduced me to an interesting discipline in which I would build my career, but also given me much support, guidance, and encouragement to become an independent researcher. I choose a teaching career because it offers me a chance to pursue my research interests, and to teach some young people the way I have been taught. I hope that some of my students will also find a lifelong fascination in part or all of computer science.