TITLE: Analysis of Randomised Search Heuristics for Dynamic Optimisation
SPEAKER: Christine Zarges (The University of Birmingham)

ABSTRACT:

Many real-world problems are subject to changing conditions over time. We refer to such problems as dynamic optimisation problems. Different randomised search heuristics are applied very successfully in this context, however, their theoretical analysis is lagging far behind practical successes. One reason for this is the absence of a commonly accepted and useful theoretical framework. In this talk I will point out important aspects that need to be considered when performing meaningful analyses in dynamic environments and introduce a novel theoretical framework based on the notion of ‘any time performance’, i.e.,  instead of concentrating on the time needed to (re-)discover a global optimum and ignoring the level of performance in the time where an optimum is not yet found or lost again, we aim at analysing the expected performance at any given point in time during the optimisation process. Afterwards I will present a case study comparing two randomised search heuristics, namely an evolutionary algorithm and an artificial immune system, in this theoretical framework using a novel dynamic example function. We will in particular discuss dynamic scenarios that take into account different parameters of the dynamic problem as well as different parameterisations of the algorithms.