Information directed sampling and ordinal optimization for combinatorial material synthesis and library design

Abstract

Combinatorial synthesis techniques have become more and more important in many areas of process and material designs. Simulated annealing is often suggested as a possible sampling policy in combinatorial methods. However, without model estimates of fitness function, true importance sampling cannot be performed. We suggested that a simple prediction model can be constructed as a generalized regression neural network using currently available. An information free energy index is defined using this model, which directs search to points that have potentially high fitness function (low information energy) and areas that are sparsely sampled (high information based entropy). Two benchmark problems were used to model the optimization problem involved in combinatorial synthesis and library design. We showed that when importance sampling is performed, the combinatorial technique becomes much more effective. The improvement in efficiency is explained using the concept of ordinal optimization.