Relatore: Prof Claudio Conti La Sapienza University and ISC-CN.

Abstract: Machine learning is percolating the way we do science. Many new computational tools are emerging and changing our approach to open problems in physics. But if we examine these tools with attention, we find out that physics mostly inspires them. This implies that we can use physical systems to implement the models in modern artificial intelligence beyond programming them in a conventional computer. This opportunity is evident in photonics and is fostering a renaissance of photonic computing. Nowadays, we can surpass the historical limitations of light-driven processors, leading to new devices at the interface of classical and quantum technologies. We show here how to realize simple but powerful optical computing machines by using spatial light modulation within an introductory approach. Specifically, we describe “Ising Machines,” which solve combinatorial optimization problems and tackle well-known large-scale systems in condensed matter physics. Also, we report on “Extreme Learning Machines,” which implement a neuromorphic computing protocol. They can be trained to act as a classifier or a universal interpolator as a conventional machine learning model. We can improve these simple computing devices in several directions, for example, by using novel materials to increase the nonlinear response or nano-fabricated surfaces to engineer specific applications while opening a new generation of energetically efficient, ultrafast, and versatile photonic computing machines.