Tuning Optimization Algorithms Under Multiple Objective Function Evaluation Budgets 

 Abstract

 Most sensitivity analysis studies of optimization algorithm control parameters are 
 restricted to a single objective function evaluation (OFE) budget. This restriction 
 is problematic because the optimality of control parameter values (CPVs) is dependent 
 not only on the problem's fitness landscape, but also on the OFE budget available to 
 explore that landscape. Therefore, the OFE budget needs to be taken into consideration 
 when performing control parameter tuning. This paper presents a new algorithm tuning 
 multiobjective particle swarm optimization (tMOPSO) for tuning the CPVs of stochastic 
 optimization algorithms under a range of OFE budget constraints. Specifically, for a 
 given problem tMOPSO aims to determine multiple groups of CPVs, each of which results 
 in optimal performance at a different OFE budget. To achieve this, the control parameter 
 tuning problem is formulated as a multiobjective optimization problem. Additionally, 
 tMOPSO uses a noise-handling strategy and CPV assessment procedure, which are specialized 
 for tuning stochastic optimization algorithms. Conducted numerical experiments provide 
 evidence that tMOPSO is effective at tuning under multiple OFE budget constraints.