Fitness inheritance for noisy evolutionary multi-objective optimization

Abstract

This paper compares the performance of anti-noise methods, particularly probabilistic and re-sampling methods, using NSGA2. It then proposes a computationally less expensive approach to counteracting noise using re-sampling and fitness inheritance. Six problems with different difficulties are used to test the methods. The results indicate that the probabilistic approach has better convergence to the Pareto optimal front, but it looses diversity quickly. However, methods based on re-sampling are more robust against noise but they are computationally very expensive to use. The proposed fitness inheritance approach is very competitive to re-sampling methods with much lower computational cost.

Evolutionary optimisation of noisy multi-objective problems using confidence-based dynamic resampling

“…It is assumed that the noise distribution is known and that the noise of all solutions comes from the same normal distribution (Bui et al, 2005). These assumptions limit the application of the technique in practice, since in real-world problems the characteristics of the noise are usually unknown and the noise level might vary in the search space (Bui et al, 2005).…”

“…These assumptions limit the application of the technique in practice, since in real-world problems the characteristics of the noise are usually unknown and the noise level might vary in the search space (Bui et al, 2005).…”

“…A technique to handle noise based on the concept of fitness inheritance has also been suggested (Bui et al, 2005). In this approach, a child inherits the mean objective valueμ = µ par ent 1 +µ par ent 2 2 and the mean standard deviationσ = σ par ent 1 +σ par ent 2 2 from its parents.…”

Surrogate-Assisted Artificial Immune Systems for Expensive Optimization Problems

“…First proposed in (Smith et al, 1995), the fitness inheritance surrogate model has been applied in several problems (Bui et al, 2005;Ducheyne et al, 2003;Salami & Hendtlass, 2003;Sastry et al, 2004;Zheng et al, 1997) and algorithms (Pilato et al, 2008;Reyes-Sierra & Coello, 2005). In this method, all the individuals in the initial population have their fitness value calculated by the exact objective function evaluator.…”

An Introduction to Multi-Objective Optimization

“…• Noisy environments: A noise effect is inevitable in many real-world problems. Sources of noise can vary depending on the way data is obtained such as the sensors, and actuators, or because of the stochastic elements pertaining in some problems such as multi-agent simulations (Bui et al, 2005a).…”