EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

4114. Scalable algorithms for throughput time constrained decision-intensive processes

Invited abstract in session MC-45: Methods and Algorithms of Decision Support, stream Decision Support Systems.

Monday, 12:30-14:00
Room: 30 (building: 324)

Authors (first author is the speaker)

1. Simon Voorberg
SISCAD, NEOMA business school

Abstract

This work is on using Reinforcement Learning for Optimal Information Acquisition in Decision Processes. We use a model-based Deep Reinforcement Learning (DRL) technique to optimize large scale decision-intensive processes. We extend a basic model for decision-intensive processes with lead time constraints and parallel information acquisition. In many situations multiple sources of information can be requested in parallel, such that we can put a maximum throughput time constraint on the process while still minimizing for the total effort that is invested in the process. These extensions cause the state space of the problem to explode and we aim at showing the value of DRL for problems like this. For small size problems we make a comparison to the optimal solution and for larger size problems we introduce some heuristics against which we compare the performance of the DRL solution. For the small size problems our solution performs within a small range from the optimal solution. For the larger case we show significant improvements compared to the heuristics.

Keywords

Status: accepted


Back to the list of papers