Researcher talks about Large Language Models Feb. 22 at 11 a.m.

Biomedical natural language processing (NLP) aims to make it easier to extract important information from unstructured texts like electronic health records, biomedical journal articles and regulatory documents, and to use this information to improve our lives.

Tim Miller will describe recent work from his Machine Learning for Medical Language Lab at 11 a.m. today in EMS E20. His talk will connect the field of biomedical NLP with the emergence of a powerful class of models known as large language models (LLMs).

He will address questions like: How important is dataset creation? Will NLP experts and subject matter experts need each other anymore? Will LLMs still suffer from out-of-domain performance loss as supervised models?

Miller is an associate professor in the Computational Health Informatics Program at Boston Children’s Hospital, the Department of Pediatrics at Harvard Medical School, and the Harvard-MIT Center for Regulatory Science. His research focuses on domain adaptation/generalizability of ML-based NLP methods, and learning patient representations.