Research news
Dissertation news

Dissertation: Dissertation: 17 Sept 2016 Algorithmic issues in Computational Intelligence Optimization: from design to implementation, from implementation to design (Caraffini)


17.9.2016 12:00 — 15:00

Location: Mattilanniemi, Agora, Alfa
PhD Fabio Caraffini defends his doctoral dissertation in Information Technology ”Algorithmic issues in Computational Intelligence Optimization: from design to implementation from implementation to design”. Opponent Prof. Daniela Zaharie, Faculty of Mathematics and Computer Science West University of Timisoara, Romania and custos Visiting professor Ferrante Neri (University of Jyväskylä).

PhD Fabio Caraffini defends his doctoral dissertation in Information Technology Algorithmic issues in Computational Intelligence Optimization: from design to implementation from implementation to design. Opponent Prof. Daniela Zaharie, Faculty of Mathematics and Computer Science West University of Timisoara, Romania and custos Visiting professor Ferrante Neri (University of Jyväskylä).

The vertiginous technological growth of the last decades has generated a variety of powerful and complex systems. By embedding within modern hardware devices sophisticated software, they allow the solution of complicated tasks. As side effect, the availability of these heterogeneous technologies results into new difficult optimization problems to be faced by researchers in the field.

In order to overcome the most common algorithmic issues, occurring in such a variety of possible scenarios, this research has gone through cherry-picked case-studies. A first research study moved from implementation to design considerations. Implementation limitations, such as memory constraints and real-time requirements, inevitably plague the algorithmic design. Such limitations are typical of embedded systems. In this light, a fast and memory-saving “compact” algorithm was designed to be used within microcontrollers. Three robotic applications were subsequently addressed by means of selected single-solution approaches and the proposed compact algorithm. A new memetic computing approach using a micro-population was also designed to tackle large scale problems. In a second moment, the opposite approach, from design to implementation, was employed. As the benefit of metaheuristic optimization is the capability of tackling black-box systems, 6 novel general-purpose optimizers were designed according to different working principles. Their validity was thoroughly tested by means of popular benchmark suites. Finally, a theoretical study concludes this piece of research. The dynamic behaviour of population-based optimization algorithms, such as Genetic Algorithm and Particle Swarm Optimization, was observed. Their general-purpose nature questioned. The presence of an intrinsic structural bias was graphically displayed and rigorously formalized. It was shown that the bias prevent them from equally exploring all the areas of the search space, with a particularly deleterious strength in presence of a large population size.

Keywords: hyper-heuristics, memetic computing, differential evolution, compact algorithms, single-solution algorithms, local search, structural bias

More information

Ferrante Neri