Christopher A. W. Vale and Kristina Shea
A method that exploits machine learning to aid modification-based computational design synthesis is presented. The algorithm makes use of statistical inference to identify appropriate modification strategies that guide the selection of modifications to develop a design solution. The procedure is especially suited to computational design tasks that cannot be easily formulated for standard algorithms. In line with the needs of most practical design tasks, the technique is oriented towards multi-objective search, yielding an archive of pareto-optimally directed design solutions. Results of application to example design tasks, truss synthesis and bitmap design, are presented.