Title: Explainable AI for Science and Medicine

Advisor: Su-In Lee

Supervisory Committee: Su-In Lee (Chair), Meliha Yetisgen (GSR, Biomedical Informatics), Ali Shojaie (Biostats), and Larry Ruzzo

Abstract: Understanding why a machine learning model makes a certain prediction can be as crucial as the prediction's accuracy in many scientific and medical applications. However, the highest accuracy for large modern datasets is often achieved by complex models that even experts struggle to interpret, such as tree-based ensembles or deep learning models. Here l will present a unified approach to explain the output of any machine learning model that was motivated by our work in medical machine learning. It connects game theory with local explanations, uniting many previous methods. By applying this approach to early-warning medical decision support we are able to use a complex, high accuracy model, and also provide explanations of the clinical risk factors that impacted the model's prediction. I will then focus specifically on tree-based models, such as random forests and gradient boosted trees, where we have developed the first polynomial time algorithm to exactly compute classic attribution values from game theory. Based on these methods we have created a new set of tools for understanding both global model structure and individual model predictions. The associated open source software supports many modern machine learning frameworks and is widely used across many industries.

 

Place: 
CSE2 271 (Gates Center)
When: 
Monday, June 3, 2019 - 12:00 to Friday, March 29, 2024 - 07:46