Joseph Perry Phillips
birth: June 30, 1967; place: Washington DC
Bachelors: Chemistry, Computer Science double major Stanford University, 1989; M.S.: Organic Chemistry, University of California Berkeley, 1991.
Science (A.I.), University of Michigan Ann Arbor, 2000.
thesis: Representation Reducing Heuristics for Semi-Automated Scientific Discovery; Advisors: John Laird and Nandit Soparkar
: (2002) DePaul University , Computer Science, Telecommunications & Information Systems
After earning his Ph.D. Dr. Phillips has an NSF postdoctoral award. He was Instructor at University of Pittsburgh, 2001-2002. In the fall of 2002, he joins the faculty of DePaul University School of Computer Science, Telecommunications & Information Systems.
Fields of Interest: Computational Scientific Discovery; Scientific knowledge discovery in databases
from Dr. Phillips' statement:
My research interests are in the small but growing field of Computational Scientific Discovery, which strives to develop algorithms and programs to help scientists create and extend scientific models. While the field has coalesced from related fields such as the philosophy of science, machine learning and knowledge discovery in databases (KDD), it must confront unique issues. Unlike the philosophy of science, we want to develop genuine artifacts to help practicing scientists. Unlike most of machine learning, we know there exists a large body of knowledge that must be consistent with any new findings and that may be used to bias search. Unlike KDD, we expect to find knowledge that is often more complex than linear-time discoverable association rules.
Computational scientific discovery seeks to extend the usefulness of computers by widening the search for the best model from a simple, quantitative question (e.g. What are the best coefficients for this equation given this equation template?) to a broader, more complex, qualitative one (e.g. What is the best system of rules that explain these phenomena?) Computers have been used to answer the first sort of question since their invention. Their usage to answer the second type of question dates back at least to the Buchanan et al MetaDendral project late 1960s.
My own research has spanned the breadth of this field. My thesis research considered practical ways that KDD could be applied to Computational Scientific Discovery , using geophysics as a case study. Since joining Buchanan's lab as a postdoc in the summer of 2000 we have developed philosophical underpinnings for scientific and medical model construction from our experience.
My current research extends this broad program. I have developed a principled method of allowing scientists to personalize heuristic functions that rank scientific models. While the functions empower scientists to define preferences that tell how well theories, laws and data may be trusted, they cannot be misused to rank clearly inferior models over superior ones. I have identified meta-knowledge that may be used to guide attribute construction (an important step to develop scientific models that posit new entities and properties) and have developed a system to automatically exploit it. I am currently developing a language for scientific knowledge and a scientific reasoning program that will use deduction and limited induction and abduction. This program could be in a stand-alone fashion or as the knowledge engine for more sophisticated visualization, discovery or other applications.
|This website was created by and is maintained
Dr. Scott Williams, Professor of Mathematics
State University of New York at Buffalo
visitors since opening 5/25/97