Robust decision-making problems by Luca Bertuccelli
Luca's broad interests are in robust decision-making problems driven by a
poorly known probabilistic process, such as robust Markov Decision Processes
with an uncertain transition model.
His work is investigating improvements
on scenario based methods to reduce the total number of simulations required
to find the worst case by using a new method called Dirichlet Sigma Point sampling. Also, he
is interested in adaptation or learning techniques that can update the
uncertainty on the transition model. In particular he is interested in improving estimator response times as well as reducing the conservatism of these
robust optimizations. Luca is also investigating robust multiple model
filtering when the transition model is not well known.
The figure shows an example of the Sigma Points (left) and a comparison of the new proposed estimator (blue) compared to a standard Maximum A Posteriori estimator (right).