A surrogate-assisted GA enabling high-throughput ML by optimal feature and discretization selection

摘要: Novel lookup-based classification approaches allow machine-learning (ML) to be performed at extremely high rates for suitable low-dimensional problems. A central aspect of such is the crucial importance placed on optimal selection features and discretized feature representations. In this work we propose study a hybrid-genetic algorithm (hGAm) approach solve optimization problem. For considered problem fitness evaluation function expensive, as it entails training ML classifier with proposed set representations, then evaluating resulting classifier. We have here devised surrogate by casting representation combinatorial in form multiple-choice quadratic knapsack (MCQKP). The orders magnitude faster allows comprehensive hGAm performance performed. results show that trade-off exists around 5000 evaluations, also provide characterization parameter behaviors input future extensions.