Machine Learning for Data-Driven Decisions (MLD3)
Welcome to the Machine Learning for Data-Driven Decisions Group at the University of Michigan!
We work at the interface of artificial intelligence (AI), machine learning (ML), and healthcare.
Our current research portfolio focuses on major public health problems – including infectious disease, Alzheimer’s disease, and diabetes, among others. We develop and apply state-of-the-art AI and machine learning methods to analyze large longitudinal health datasets. Our work spans several aspects of AI including time-series analysis, reinforcement learning, computer vision, and causal inference. We aim to develop the computational methods needed to help organize, process, and transform data into actionable knowledge with the ultimate goal of improving health.
You can contact the group by emailing Dr. Jenna Wiens at email@example.com
Two papers from our lab are accepted at MLHC 2020! Congratulations to students Ian Fox (Deep RL for blood glucose control) and Sarah Jabbour (exploiting & preventing shortcuts in Deep Learning applied to Chest X-Rays)!
Our paper on clinician-in-the-loop reinforcement learning with near-optimal set-valued policies is accepted at ICML 2020! Congratulations to students Shengpu Tang and Aditya Modi!
Forthcoming paper in the Journal of the American Medical Informatics Association by Shengpu Tang et al.
Tang et al. developed FIDDLE, a tool to help researchers preprocess structured clinical data in a systematic and reproducible manner. FIDDLE was validated on MIMIC-III and the eICU Collaborative Research Database and shown to be effective in extracting features for predicting various clinically relevant outcomes — in-hospital mortality, acute respiratory failure, and shock. Read more –>
Deep Learning Applied to Chest X-Rays: Exploiting and Preventing Shortcuts | MLHC 2020
Deep Reinforcement Learning for Closed-Loop Blood Glucose Control | MLHC 2020
Friday Night AI:
AI and COVID-19
How can machine learning impact healthcare?
Prof. Jenna Wiens uses machine learning to make sense of the immense amount of patient data generated by modern hospitals. This can help alleviate physician shortages, physician burnout, and the prevalence of medical errors.