Physics Colloquium | Machine Learning Models for Time Series Forecasting and Question Answering on Medical Data, Jan. 31

The Physics & Astronomy Colloquium Series presents Razvan Bunescu of Russ College of Engineering and Technology, Ohio University on “Machine Learning Models for Time Series Forecasting and Question Answering on Medical Data”, on Friday, Jan. 31, at 4:10 p.m. in Clippinger Labs 194.

Razvan Bunescu

Abstract: Two major applications of machine learning come from data acquired on individuals with type I diabetes. Avoiding serious complications in type I diabetes requires good blood glucose control. If blood glucose levels could be accurately predicted, patients could take proactive steps to prevent blood glucose excursions.

First, a recursive neural network approach that uses long short-term memory (LSTM) units to learn a physiological model of blood glucose is presented. When trained on raw data from real patients, the LSTM networks obtain results that are competitive with human experts and with a previous model based on manually engineered physiological equations.

Second, a neural sequence-to-sequence model is trained using reinforcement learning to automatically answer queries from doctors interested in understanding the state of a patient.