BEGIN:VCALENDAR BEGIN:VEVENT SUMMARY:PhD Defense - Greg Harris DESCRIPTION:Customized Data Mining Objective Functions\n \n Ph.D. candidate: Greg Harris\n \n Wednesday, Nov. 30, 2016\n 10:30 AM, EEB 110\n \n \n Abstract:\n Interpretable machine learning models, such as classification rule lists, enable knowledge discovery and model vetting by domain experts. Their transparency, however, often comes at the cost of accuracy, when compared to more complex models. Our research seeks to improve the accuracy of such models while retaining their interpretable rule-based form. Our strategy is to generate domain-dependent objective functions that specify heuristic trade-offs tailored for individual datasets.\n \n Our first contribution is FrontierMiner, a new rule-based algorithm for predicting a target class with high precision. It learns a non-parametric objective function directly from the data. We show that FrontierMiner finds higher-precision rules more often than competing rule induction systems in a study involving 1,000 synthetic datasets and 138 real-world classification tasks. Our second contribution is PRIMER, a new algorithm for maximizing event impact on time series. It has an objective function that adapts to the level of noise in the data. It also incorporates user-provided input on the expected response pattern as a heuristic that helps prevent over-fitting. We show PRIMER is competitive with state-of-the-art regression techniques in a large financial event study, yet has improved model interpretability. Our third contribution is a method of learning an objective function from user feedback in the form of pairwise rankings. With this feedback, we use learning-to-rank algorithms to combine existing measures into an overall objective function that more closely matches the user's preference. We conclude the presentation with directions for future research.\n \n \n Biography: \n Greg Harris is currently a PhD candidate in the Computer Science Department at the University of Southern California. His research interests include data mining, pattern recognition, and machine learning. He also holds a Master of Financial Mathematics degree from the University of Minnesota and a Bachelor of Science degree in Applied Physics from Brigham Young University. \n \n \n Defense Committee: Viktor Prasanna (chair), Cauligi Raghavendra, Ellis Horowitz\n \n \n \n \n DTSTART:20161130T103000 LOCATION:EEB 110 URL;VALUE=URI: DTEND:20161130T123000 END:VEVENT END:VCALENDAR