back

Fitness landscape characterisation for search-based software engineering problems

Seminar


Tuesday, June 17, 2014 at 10:00 – GSSI Room B

Dr. Aldeida Aleti

 

Abstract

The design and development of today’s software systems is a complex task, involving several design decisions, which have significant implications for the quality of the final system. The ever-increasing complexity of software systems introduces a big challenge for system engineers. They have to choose from a growing number of design options and a design space that is beyond the capability of humans to understand.

 

To handle this complex task, optimisation methods that automate the search for optimal software architectures have been introduced. Different optimisation methods may be suitable for different problems. The suitability of an optimisation method is closely related to the structure of the fitness landscape. For instance, a local search method performs well in unimodel fitness landscapes, whereas genetic algorithms are more suitable for rugged, multimodal fitness landscapes. In this talk, I will introduce some methods for measuring the structure of fitness landscapes, and determine the suitability of various optimisation methods for solving problems in the software engineering domain.