An optimization algorithm for imprecise multi-objective problem functions

Real world objective functions often produce two types of uncertain output: Noise and imprecision. While there is a distinct difference between both types, most optimization algorithms treat them the same. This paper introduces an alternative way to handle imprecise, interval-valued objective functions, namely imprecision-propagating MOEAs. Hypervolume metrics and imprecision measures are extended to imprecise Pareto sets. The performance of the new approach is experimentally compared to a standard distribution-assuming MOEA.