Feedforward neural networks in catalysis: A tool for the approximation of the dependency of yield on catalyst composition, and for knowledge extraction

Artificial neural networks are distributed computing systems implementing the functionality characterizing biological neural networks. This way of computing has become quite successful in practical applications as a tool for solving several traditional mathematical and data-analysis tasks, such as classification, clustering, approximation and prediction. In this paper, main principles of employing multilayer perceptrons for the approximation of unknown functions are outlined, and another possible use of multilayer perceptrons in combinatorial catalysis is indicated—their use for the extraction of knowledge from experimental catalytic input and output data. To counterbalance the abstractness of the subject, the method is illustrated by applying multilayer perceptrons to data on catalyst composition and catalytic results in the oxidative dehydrogenation of propane to propene.