With a study published in the prestigious journal Optica, researchers in the Physics Department at the Politecnico di Milano have built a connection between two fields: artificial intelligence, which has been increasingly studied in recent years, and non-linear optics.
The research, conducted by Carlo Michele Valensise, Giulio Cerullo, and Dario Polli, together with Alessandro Giuseppi from Sapienza Università di Roma, began about a year ago during the first lockdown. It is based on the study of Deep Reinforcement Learning (DRL), that is, the branch of artificial intelligence related to programming agents that can learn to control automated systems. In other words, a DRL agent “learns” thanks to the independent interaction with the system in front of it.
Laboratory experiments then confirmed that the application of DRL to non-linear optics allows for simplification of some processes and, more generally, to speed up experimentation. One possible application, for example, is found in the generation of white light, one of the most common phenomena in this field of research.