During your last visit to your doctor, how much of your doctor’s attention was focused on you, and how much on her computer?

Doctors cope with the burden of processing more and more data, mostly in the form of free text.  NLP algorithms can alleviate this burden. However, NLP algorithms in the Healthcare domain have their own set of challenges: the scarcity of training due to privacy concerns, the lack of standardization, and the multilingual nature of medical data, all present a challenge for classic transformer-based NLP Models.

This lecture will discuss how those challenges were solved at Microsoft while working on Azure Text Analytics for Healthcare,  a service that provides deep analysis of multilingual medical texts. The lecture explores Federated Learning algorithms and how they were used  to solve data privacy issues, as well as the XLM-K Language Model and how it was used to train the model on seven languages with almost no training data. Finally, this lecture explains how medical ontologies (UMLS and others) can fill gaps in medical knowledge.

This talk is intended for women interested in the medical data science domain, as well as for data scientists coping with similar problems in other domains