Project Details
Projekt
Optimal actions and stopping in sequential learning
Subject Area
Mathematics
Term
since 2021
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 460867398
In this project, we consider the general problem of sequential learning. We study it in the problem of adapting the sampling strategy for linear regression as well as in the case of the Partial Least Squares (PLS) algorithm. A fundamental problem for the PLS algorithm is to decide when it should be stopped, i.e.~to solve a stopping problem. Another important problem, in the case where multiple instances of such an algorithm are run in parallel, is to determine how the computational budget should be allocated between the different problems, i.e. solve a sequential resource allocation problem. A central difficulty in both these problems is related to uncertainty quantification: both for optimal stopping, and for optimal resource allocation, it is necessary to produce reliable estimators of the uncertainty achieved by PLS at all times. This can be quite difficult in the case of high dimensional regression under complex constraints. We aim in this project at overcoming this problem, and at proposing clear guarantees for PLS.
DFG Programme
Research Units
