Events and Seminars : 2013 Seminars

Mining for Rules in Data:  Modeling Imperfect Implication Rules

Kamal Premaratne, Ph.D.
Professor, Dept of Electrical and Computer Engineering, University of Miami

February 14, 2013
2:00-3:00 p.m, CRB Room 692

Rule mining involves the extraction of rules of the type “if A then B” from data. Rules extracted from a finite set of (potentially imperfect) data can hardly be considered perfect. Expert opinions, which are often expressed in terms of natural language statements, tend to generate rules of the type “if A then B with a confidence of 70%”, or more likely, “if A then B with a confidence between 70% and 80%”. In reality, the situation is even more complicated because of the uncertainty or ambiguity regarding the occurrence of the rule antecedent A (e.g., . “I am 75% confident that A occurred”). Such imperfect implication rules are central to any process of reasoning because they capture how humans express knowledge and uncertainty. How can we model imperfect implication rules and use them for extraction of knowledge? The Bayesian approach does not appear to be well equipped to handle imperfect implication rules. In this talk, we provide a brief introduction to Dempster-Shafer (DS) belief theory and show how it can provide an effective framework for modeling imperfect implication rules, and in general imperfect logic constructs.