On the design, analysis, and characterization of materials using computational neural networks

Since the resurgence of computational neural networks (CNNs) (about 10 years ago with the popularization of backpropagation), almost all scientific and technical fields have made use of them in some form, often reporting surprising advantages. Although it is not clear whether CNNs are truly an emerging technology or just a subset of other fields, it is clear that CNNs do provide useful characteristics suitable for a broad range of applications. A diverse set of problems in materials science have enjoyed the flexibility and power that is offered by CNNs. Applications include making structure-activity/property relationships; predicting chemical reactivity; process control; modeling, optimization, and diagnosis; pattern recognition and classification of spectra; and data analysis, to name a few. Such diversity stems from the fact that CNNs provide a general and tractable tool for problem solving. In this article we review the basic elements of CNNs and how this computational technique has been applied in materials science.