education

Finding structure in P+O loss functions

Thomas Kuiper (2024)

A common trait of loss functions in the predict-and-optimize field is that they give up some information about the problem to make the loss function continuous. To get a better idea of where the strengths and weaknesses lie of P+O losses, we want to find how much of this information on problem structure is retained. In this project, we would like to address the following question: How much of the problem structure do P+O loss functions capture?

Decision Diagram Focused Learning: Efficient Predict-and-Optimize With Decision Diagrams

Jop Schaap (2024)

Co-supervised with Koos van der Linden

An issue present in many state-of-the-art approaches to predict-and-optimize is the need to re-solve the same problem with different objective coefficients for each gradient computation. In this project, we investigate strategies for predict-and-optimize based on decision diagrams as a representation of a feasible set; since changing the objective values corresponds to re-weighting the diagram, our hypothesis is that we can derive a faster learning routine based on this idea.

Combinatorial Optimisation for Scheduling (2023)

In this project, we have investigated the strategies for guiding a SAT solver into better solution for several variations of resource-constrained project-scheduling problem (RCPSP). A common theme between the subprojects was the importance of the seed values of the VSIDS branching heuristics; despite the common approval of VSIDS as a branching strategy, it has no built-in solution for the cold-start issue. The results of this project suggest several strategies for initializing VSIDS values in a way that triggers a solver into common heuristic solutions.

• Augmenting VSIDS heuristic for the RCPSP/t by initializing activity values using domain-specific information
• Integration of variable selection heuristics into a MaxSAT solver for solving the multi-mode resource-constrained project scheduling problem
• Combinatorial optimization for job sequencing with one common and multiple secondary resources by using a SAT solver augmented with a domain-specific heuristic
• Combining SAT solvers with heuristic ideas for solving RCPSP with logical constraints
• Why Midas would be a terrible secretary