QUERY-DRIVEN EXPLORATION OF MODEL PREDICTIONS

Deep learning has transformed the way that we analyze image data. Can it do the same for other data types? How to we bring scientists 'into the loop', so that they can understand decisions made by neural networks?

In collaboration with Argonne National Laboratory, I am developing visualization approaches to allow researchers to explore machine learning outputs on the CANDLE project, which develops deep learning models to predict drug treatment outcomes in precision medicine. My goal is to enable researchers to identify subsets of the data with unusual prediction results, to understand model outputs.

This work is in progress and will be a significant component of my thesis work.

Data science researchers at Argonne National Laboratory: Rick Stevens, Fangfang Xia, and Alexander Partin.