Landscape Analysis Under Measurement Error

Abstract

There are situations where the need for optimisation with a global precision tolerance arises — for example, due to measurement, numerical or evaluation errors in the objective function. In such situations, a global tolerance ε > 0 can be predefined such that two objective values are declared equal if the absolute difference between them is less than or equal to ε. This paper presents an overview of fitness landscape analysis under such conditions. We describe the formulation of common landscape categories in the presence of a global precision tolerance. We then proceed by dis- cussing issues that can emerge as a result of using tolerance, such as the increase in the neutrality of the fitness landscape. To this end, we propose two methods to exhaustively explore plateaus in such application domains — one of which is point-based and the other of which is set-based.