Reinforcement Learning
joint work with S. Allassonnière, R. Besson, F. Logé-Munerel, T. Levent, P. Clavier, H. Castel, E. Hyon and O. Forghieri
With S. Allassonnière, R. Besson, we have worked on an optimization of a decision tree to accelerate foetal abnomalies detection. This tools combines an expert model with data assimilation and a reinforcement learning procedure on trees. It is now used by Sonio.
With F. Logé-Munerel and Air Liquide, we have used a similar idea to optimize health questionnaire and applied reinforcement learning to control insulin pumps.
With T. Levent, we have shown how to use reinforcement learning to optimize an energy mix.
With S. Allassonnière and P. Clavier, we are working on robust versions of reinforcement learning and their application to patient trajectories.
With H. Castel, E. Hyon and O. Forghieri, we are studying adaptive state aggregations to speed up MDP solving.