Dartmouth Events

Abstract: Large language models stand to make healthcare cheaper, faster, and more accessible. But despite gaining popularity, their responsible adoption at scale remains limited by vast challenges. In this talk, I will discuss two such technical challenges and introduce some of my recent work towards their solution. First, I will discuss social biases in training data and how we can leverage large language models to detect and mitigate implicitly harmful natural language. Second, I will discuss lifelong model editing, a new path towards real-time updates for large, expensively-trained models to quickly correct their failures without costly retraining.

Bio: Tom Hartvigsen is an Assistant Professor of Data Science at the University of Virginia. He works to make machine learning trustworthy, robust, and socially-responsible enough for deployment in high-stakes, dynamic healthcare settings. Tom focuses on NLP and time series, publishing at the top venues in Machine Learning, NLP, and Data Mining including NeurIPS, ACL, KDD, and AAAI. Before joining UVA, Tom was a postdoc at MIT CSAIL working with Marzyeh Ghassemi. He holds a Ph.D. in Data Science from WPI and a bachelor’s in Applied Math from SUNY Geneseo.