Mikita Shchutski, Lead Machine Learning Engineer, Quantori.
The talk is about training large medical models on electronic health records (EHR) in order to build a highly informative medical embedding space. Its three parts: adapting EHR for language modeling, adapting BERT to EHR (following the BEHRT paper), and what it would take to get to "large medical models". Most of the talk is about the data, because that is where most of the problems are.
Continue reading — it's free
Google or a one-time email code. No password.