Deep learning procedures for the inverse design of electromagnetic devices

Artificial intelligence (AI) has revealed the most investigated topic in the last years in many branches of scientific research, ranging from medicine to civil engineering and ICT applications. In this framework, different problem-solving paradigms emerged during the years related to the most convenient way of dealing with the available data from which the AI has to take profit. In the field of applied electromagnetics, deep learning (DL) procedures have been extensively investigated so far. Actually, they revealed effective in dealing with situation of practical interest (e.g., biomedical diagnostics) because of their capability of learning the relevant features from the available data. Hence, in developing this kind of tools a key role is played by the training dataset, as well as by the user’s experience in setting the network parameters in a proper fashion.