A generic and computationally efficient automated innovization method for power-law design rules

Automated Innovization (AutoI) originally aimed to extract power-law based rules from a design optimization task without any human intervention. Existing AutoI methods have employed Evolutionary Algorithms (EAs) twice: first to search for Pareto-optimal (PO) solutions, and second to extract rules hidden in them. Furthermore, these methods are limited in scope, in that, they are not capable of tackling both continuous and discrete variables for either single-cluster or multi-cluster rules. In a unique departure from the state-of-the-art, this paper presents a computationally efficient, single EA-based, AutoI method capable of simultaneously extracting single-cluster or multi-cluster rules for both continuous and discrete variable spaces. The robustness of the proposed method is evident from its successful rule extraction tasks even with small-size datasets, where existing AutoI methods fail to apply. The generic scope, computational efficiency, and robustness of the proposed method are demonstrated through a number of benchmark design problems.