Inhalt des Dokuments
|Project heads:||Ralf Borndörfer
, Guillaume Sagnol |
|Researcher:||Daniel Schmidt gen.
This research is is supported by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy-- The Berlin Mathematics Research Center MATH+ (EXC-2046/1, project ID: 390685689).
Scheduling jobs of stochastic duration is ideally done by means of fully adaptive policies. In applications such as surgery scheduling, however, highly volatile regimes cannot be implemented. We will study appropriate scheduling policies for such situations, both from a theoretical and a computational perspective.
We are conducting a fine-grained analysis of the adaptivity trade-off for machine scheduling problems, by considering classes of policies that can be described with a single adaptivity parameter, and that interpolates fully adaptive non-anticipative policies and fixed assignment policies. The goal is to get the most out of the performance of adaptive policies,while keeping the stability properties of fixed assignments. Such policies could easily be implemented in stochastic environments like hospitals.
From a more practical point of view, we are also investigating the use of reinforcement learning to compute such policies.