We are continuing hiring: https://lnkd.in/gyZQzPHT.
We're looking for aspirational candidates with strong research/engineering experience to join our quest in advancing #GenAI for precision health.
The confluence of digital transformation in medicine and GenAI revolution enables pretraining of powerful biomedical frontier models, which can synthesize multimodal, longitudinal patient journey to create a holistic representation as the digital twin for the patient. Just like GPT4 predicts next token, such a high-fidelty "patient embedding" can predict next major medical events such as disease progression and treatment response. It can instantly unlock patient-like-me reasoning at the population scale, democratizing high-quality patient care by identifying what works and accelerating discovery by pinpointing exactly where and how current targets and drugs fail.
This is exciting, but today's GenAI is not up for the task yet. Frontier models still suffer from hallucination and omission errors. Real-world data is plagued by confounders. State-of-the-art large multimodal models (LMMs) barely scratch the surface of multimodal complexity in biomedicine.
At Microsoft Health Futures, we conduct a wide range of innovative research in GenAI for precision health, such as self-improving LLM agent, causal real-world evidence, multimodal learning. Microsoft Research at large features amazing interdisciplinary groups such as Deep Learning (Jianfeng Gao), AI4S (Tao Qin), MSRA (Furu Wei), Biomedical Imaging (Javier Alvarez Valle), Biomedical Signal Processing (Michael Hansen), Biomedical ML (Nicolo Fusi), AI Frontiers (Ece Kamar). We are fortunate to collaborate with many of these teams, as well as Nuance Communications, GenAI, and external collaborators such as academia, health systems and life science companies.
For more info about our research & incubation, check out some relevant pointers:
KDD tutorials: https://lnkd.in/gVrsG7Qu
NEJM AI position paper: https://lnkd.in/gtubhGVT
Microsoft Research Forum briefing: https://lnkd.in/gheD5Svn
