Organic material that learns

Construction of physical neural networks for reservoir computing

Abstract

The possibility of a neural network structure utilizing any response from devices, materials or physical substances, for which a function can be used for information processing, is now attracting considerable attention. The realization of a neural network consisting of materials requires experimental processes and knowledge that might not be common in the research field of applied physics. We demonstrated the structure of two different types of neural networks utilizing the plasticity of polymer growth and the nonlinear response of the electrochemical reaction of molecules. Our organic neural networks show primitive functionality for information processing. In this paper, the system of basic neural network and reservoir computing are illustrated and explained in plain words, and experimental processes, namely, how the materials learn to be a neural network and how we evaluate their performance, are concretely explained.