Vous êtes ici : Accueil > Université > Composantes > Faculté d’Informatique > Séminaires IRIT UT Capitole
- Recherche,
Machine Learning for Public Health Decision Making
le 11 mars 2024
12h45
Manufacture des Tabacs
Bâtiment F (Salle MF103)
Bâtiment F (Salle MF103)
Bryan Wilder is an Assistant Professor in the Machine Learning Department at Carnegie Mellon University. His research focuses on AI for equitable, data-driven decision making in high-stakes social settings, integrating methods from machine learning and optimization. Much of this work is motivated by public health applications, including maternal and child health, Covid-19, and HIV prevention. His research has been recognized with awards including a Schmidt AI2050 Early Career Fellowship, the IFAAMAS Victor Lesser Distinguished Dissertation Award, and best paper nominations at ICML and AAMAS.
Abstract:
Decision making in high-stakes public health settings often requires allocating limited resources, for example the availability of community health workers or limited supplies of a therapeutic. This talk introduces machine learning methods both to improve such allocations and to assess the efficacy and fairness of allocations made by either humans or algorithmic systems. We start by introducing a framework for decision-focused learning, where we incorporate an optimization problem modeling constrained decision making into the training loop of a machine learning model which predicts unknowns appearing in the problem (e.g., the risk level of each patient). Then, we will discuss the challenges of assessing the real-world performance of both algorithmic and human systems for making such decisions, and introduce improved statistical methods which offer greater robustness to challenges such as confounding and correlations in treatment assignment.
Decision making in high-stakes public health settings often requires allocating limited resources, for example the availability of community health workers or limited supplies of a therapeutic. This talk introduces machine learning methods both to improve such allocations and to assess the efficacy and fairness of allocations made by either humans or algorithmic systems. We start by introducing a framework for decision-focused learning, where we incorporate an optimization problem modeling constrained decision making into the training loop of a machine learning model which predicts unknowns appearing in the problem (e.g., the risk level of each patient). Then, we will discuss the challenges of assessing the real-world performance of both algorithmic and human systems for making such decisions, and introduce improved statistical methods which offer greater robustness to challenges such as confounding and correlations in treatment assignment.
Mis à jour le 26 janvier 2024
