home · mobile · calendar · colloquia · 2000-2001 · 

Colloquium - Margineantu

Learning with Costs
Department of Computer Science, Oregon State University

Cost-sensitive learning addresses the task of learning to make predictions and decisions when different types of errors have different associated costs (penalties). Existing learning algorithms were designed to treat all errors the same and to minimize the number of errors that are made. In most practical applications, however, different errors have different costs, and methods are required that can minimize the total cost of all errors. For example, in medical diagnosis, the cost of diagnosing a patient as healthy when he or she has a life-threatening disease is much higher than the cost of making the opposite mistake.

The talk will first outline the issues that need to be considered in designing and evaluating learning methods for the minimization of total cost. Then I will review the current approaches to constructing learning algorithms in a cost-sensitive context. Finally, I will introduce new methods for building and evaluating cost-sensitive learning systems and present experiments that compare different cost-sensitive methods on both real-world and synthetic data sets.

Hosted by Michael Mozer and James Martin.
Refreshments will be served immediately following the talk in ECOT 831.

Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:13)