2019/04/12 by George T. Hall, Pietro S. Oliveto, Hall, George T. +3
Biochemistry, Genetics and Molecular Biology · Computer Science · #Advanced Multi-Objective Optimization Algorithms #Evolutionary Algorithms and Applications #FOS: Computer and information sciences #Metaheuristic Optimization Algorithms Research #Neural and Evolutionary Computing (cs.NE) #vaccines and immunoinformatics approaches
paper · pdf · doi:10.48550/arxiv.1904.06230
openalex publication_date 2019/04/12 · openalex created_date 2022/07/29 · openalex updated_date 2026/07/28
Algorithm configurators are automated methods to optimise the parameters of\nan algorithm for a class of problems. We evaluate the performance of a simple\nrandom local search configurator (ParamRLS) for tuning the neighbourhood size\nk of the RLSk algorithm. We measure performance as the expected number of\nconfiguration evaluations required to identify the optimal value for the\nparameter. We analyse the impact of the cutoff time \κ (the time spent\nevaluating a configuration for a problem instance) on the expected number of\nconfiguration evaluations required to find the optimal parameter value, where\nwe compare configurations using either best found fitness values (ParamRLS-F)\nor optimisation times (ParamRLS-T). We consider tuning RLSk for a variant of\nthe Ridge function class (Ridge*), where the performance of each parameter\nvalue does not change during the run, and for the OneMax function class, where\nlonger runs favour smaller k. We rigorously prove that ParamRLS-F efficiently\ntunes RLSk for Ridge* for any \κ while ParamRLS-T requires at least\nquadratic \κ. For OneMax ParamRLS-F identifies k=1 as optimal with\nlinear \κ while ParamRLS-T requires a \κ of at least \Ω(n\log\nn). For smaller \κ ParamRLS-F identifies that k>1 performs better\nwhile ParamRLS-T returns k chosen uniformly at random.\n