GLYMOUR@C.CS.CMU.EDU (Clark Glymour) (07/17/87)
SEMINAR IN LOGIC AND COMPUTABILITY: ARTIFICIAL INTELLIGENCE AND FORMAL LEARNING THEORY - Offered by: Department of Philosophy, Carnegie-Mellon University - Instructor: Kevin T. Kelly - Graduate Listing: 80-812 - Undergraduate Listing: 80-510 - Place: Baker Hall 131-A - Time: Wednesday, 1:30 to 4:30 (but probably not the full period). - Intended Audience: Graduate students and sophisticated undergraduates interested in inductive methods, the philosophy of science, mathematical logic, statistics, computer science, artificial intelligence, and cognitive science. - Prerequisites: A good working knowledge of mathematical logic and computation theory. - Course Focus: Convergent realism is the philosophical thesis that the point of inquiry is to converge (in some sense) to the truth (or to something like it). Formal learning theory is a growing body of precise results concerning the possible circumstances under which this ideal is attainable. The basic idea was developed by Hilary Putnam in the early 1960's, and was extended to questions in theoretical linguistics by E. Mark Gold. The main text of the seminar will be Osherson and Weinstein's recent book Systems that Learn. But we will also examine more recent efforts by Osherson, Weinstein, Glymour and Kelly to apply the theory to the inductive inference of theories expressed in logical languages. From this general standpoint, we will move to more detailed projects such as the recent results of Valiant, Pitt, and Kearns on polynomial learnabilitly. Finally, we will examine the extent to which formal learning theory can assist in the demonstrable improvement of learning systems published in the A.I. machine learning literature. There is ample opportunity to break new ground here. Thesis topics abound. - Course Format: Several introductory lectures, Seminar reports, and a novel research project. PROBABILITY AND ARTIFICIAL INTELLIGENCE - Offered by: Department of Philosophy, Carnegie-Mellon University - Instructor: Kevin T. Kelly - Graduate Course Number: 80-312 - Undergraduate Course Number: 80-811 - Place: Porter Hall, 126-B - Time: Tuesday, Thursday, 3:00-4:20 - Intended Audience: Graduate students and sophisticated undergraduates interested in inductive methods, the philosophy of science, mathematical logic, statistics, computer science, artificial intelligence, and cognitive science. - Prerequisites: Familiarity with mathematical logic, computation, and probability theory - Course Focus: There are several ways in which the combined system of a rational agent and its environment can be stochastic. The agent's hypotheses may make claims about probabilities, the agent's environment may be stochastic, and the agent itself may be stochastic, in any combination. In this course, we will examine efforts to study computational agents in each of these situations. The aim will be to assess particular computational proposals from the point of view of logic and probability theory. Example topics are Bayesian systems, Dempster-Shafer theory, medical expert systems, computationally tractable learnability, automated linear causal modelling, and Osherson and Weinstein's results concerning limitations on effective Bayesians. - Course Format: The grade will be based on frequent exercises and possibly a final project. There will be no examinations if the class keeps up with the material.