Title: The strength to say “I Don't Know”: expanding neural networks to address out-of-distribution samples and ambiguous inputs 

Abstract: Machine Learning methods, and specifically neural networks, have shown impressive capabilities in a wide range of application areas.  However, these models are often brittle when seeing unexpected or novel data that differ in meaningful ways from the data that was used to train them.   Commonly, this is overcome by ever-increasing training data volumes.  However, there are many applications that are inherently data-limited or scenarios in which systems encounter ambiguous or out-of-distribution samples despite massive volumes of training data.  In these cases, the ability of a model to be able to produce the output “I don’t know, never seen anything like this before” can be critical. In this talk, I will discuss methods that my group has developed to flag out-of-distribution and ambiguous inputs in test data using a Null Space Analysis of network layers and by incorporating novel network feature extraction layers.

Bio: Alina Zare teaches and conducts research in the area of machine learning and artificial intelligence as a Professor in the Electrical and Computer Engineering Department at the University of Florida. She also serves as the Associate Dean for Research and Facilities at the Herbert Wertheim College of Engineering at the University of Florida.  Dr. Zare’s research has focused primarily on developing new machine learning algorithms to automatically understand and process data and imagery. Her research work has included automated plant root phenotyping, sub-pixel hyperspectral image analysis, target detection, and underwater scene understanding using synthetic aperture sonar, LIDAR data analysis, Ground Penetrating Radar analysis, and buried landmine and explosive hazard detection.