Hyper & Meta-Heuristics Design for Combinatorial Optimization

Many challenging applications in Science and Industry can be formulated as optimization problems. Due to their complexity and hardness often they cannot be solved in an exact manner within a reasonable time; therefore, approximate algorithms become the main solving alternatives  thanks to their ability to efficiently explore large search spaces.

Metaheuristics are successful techniques able to solve such complex, and hard optimization problems that arise in human activities, such as economics, industry, or engineering, and constitute a highly diverse family of optimization algorithms, each of which shows individual properties, and different strengths.

However, the main limitation, primarily for the standard optimization methods, is to produce the best results for a few problem instances, while often it performs poorly on the other problem instances. The Hyper-heuristics methods instead attempt to provide more generalized solutions to combinatorial optimization problems by working in the "heuristic space" rather than the "solution space". Basically, they select or generate "low-level" heuristics for solving the problem at hand, attempt to combine the strengths of some heuristics to compensate for the weaknesses of the others.

The aim of the project is to design and develop hyper-heuristics systems and /or metaheuristics able to solve classical combinatorial problems, as well as any application of important relevance in the real life.

Tutor of the traineeship: prof. Mario Pavone