Statistics and Statistical Learning Seminar: EMPIRICAL RISK MINIMIZATION OF AUROC
Tuesday, February 18, 2020 - 3:00pm
Tuesday February 18th, at 3:00 pm, in FMH-462 (large conference room)
speaker:  Victor Rielly

For unbalanced binary classification problems, or problems requiring a trade off between true positive and false positive rates, machine learning models are often evaluated using AUROC (Area under the Reciever Operating Characteristic Curve). Many papers have been written in an attempt to create machine learning models that optimize directly for AUROC. We present a review of the current literature on the subject, some new results and interpretations of the task, as well as some experimental results.

Victor received a bachelor's degree in Mathematics and Physics with a minor in Computer Science from Pacific University. He has been involved in projects ranging from the fields of Chemistry (counting orientations of molecules using chromatic polynomials on graphs and automorphism groups on graphs), biology (doing preliminary tests of Electroencephalogram and galvanic skin response sensors for a crew state monitoring system project), to rocket science (building a test suite for a software system designed to test NASA's Space Launch System), physics (designing an algorithm that can be used to sort light signals for the purpose of developing light based computing) to mathematics (Billiards in hyperbolic space, graph theory and topology, differential equations, and machine learning), and computer science (building a website for Nike employees, and developing an internet of things device for a company in Hillsboro).