Clinical NLP and Deep Learning… It’s Complicated

When: Wednesday, March 6, 2019
Where: PGH 232
Time: 11:00 AM

Speaker: Dr. Kirk Roberts, UTHealth School of Biomedical Informatics

Host: Dr. Thamar Solorio

Deep learning has revolutionized many areas of artificial intelligence, including natural language processing (NLP). But when applying NLP to important data such as clinical data in an electronic health record (EHR), deep learning’s many strengths start to become liabilities. The lack of data, unpredictable generalizability, and limited interpretability create numerous challenges in applying state-of-the-art deep learning methods to clinical NLP tasks. This talk will start with a broad overview of NLP for health and medicine. It will then drill down to cover some of our attempts to apply deep learning to clinical NLP, the challenges we’ve run into, and the lessons we’ve learned in overcoming them.

Bio:

Kirk Roberts is an assistant professor at the UTHealth School of Biomedical Informatics in Houston. His research interests include both natural language processing (NLP) and information retrieval (IR), particularly how they can be applied within biomedical informatics. He has organized six shared task competitions, including the recent TREC Precision Medicine track, and is an organizer of the Clinical NLP workshop series. He is also a recipient of the NIH Career Development Award.