Early applications of Neural Networks to plasma science: Architectures, solutions, and impact.

Abstract

Many applications of Neural Networks (NN) to plasma science have appeared in the last years. The author describes here some of the early applications of NNs to plasma science at the beginning of the 90 s, when multi-layer, feed-forward-back-propagation (FFBP) architectures found several applications in this field: they were used to solve inversion problems, to create complete sets of input data, to replace time-consuming modules in models and to predict the outcome of real processes. From a partially personal perspective the author reviews the details of plasma problems to which NNs were successfully applied, and those of the related architectures. It turns out that some solutions, which are perceived today as marking the difference between the previous and contemporary NNs application practices, were in common use >30 years ago when they were deemed fruitful. This can help create deeper historical insight into a field that is getting much attention today.

Keywords

• Neural networks
• Plasma modeling
• Cross sections
• Rate coefficients
• Embedded AI