Michela Milano (University of Bologna)
Empirical Model Learning: Boosting Optimization Through Machine Learning
One of the biggest challenges in the design of decision support and optimization tools for complex, real-world, systems is coming up with a good combinatorial model. The traditional way to craft a combinatorial model is through interaction with domain experts: this approach provides model components (objective functions, constraints), but with limited accuracy guarantees. Often enough, accurate predictive models (e.g. simulators) can be devised, but they are too complex or too slow to be employed in combinatorial optimization.
In this talk, we propose a methodology called Empirical Model Learning (EML) that relies on Machine Learning for obtaining decision model components that link decision variables and observables, using data either extracted from a predictive model or harvested from a real system. We show how to ground EML on a case study of thermal-aware workload allocation and scheduling. We show how to encapsulate different machine learning models in a number of optimization techniques.
We demonstrate the effectiveness of the EML approach by comparing our results with those obtained using expert-designed models.
Thorsten Koch (Zuse Institute Berlin)
Same, Same, but Different: A mostly Discrete Tour through Optimization
This talk will give a short tour through selected topics in mathematical optimization. Though these topics are quite diverse, they also have a lot in common.
The tour will start at mixed-integer non-linear optimization (MINLP), proceed to mixed-integer optimization (MILP), it will then make short detour to linear programming (LP) and exact solutions, then proceed to algorithms, software, modelling, and parallel computing, jumping to gas networks as an application, from there visit Steiner tree problems, and finally arrive back at MILP.
On route, we will take the opportunity to point out a few challenges and open problems.
Paul Shaw (IBM)
Ten Years of CP Optimizer
CP Optimizer is the IBM constraint solving engine and part of CPLEX Optimization Studio. This talk takes a look at both the motivation and history of CP Optimizer, and the ten year journey from its beginnings until today.
At selected points, I will delve into the operation of different features of the engine, and the motivation behind them, together with how performance improvements in the automatic search were achieved.
From more recent history, I will concentrate on important developments such as the CP Optimizer file format, presolve, explanations for insolubility and backtrack, and lower bounds on the objective function.