Mirco Milletari

À propos

I am currently working in the field of Machine Learning and Neural Networks. I work both on applied and fundamental problems. In the former case, I apply ML methods for recommender systems, image recognition and Natural Language processing. In the latter, I work on foundational aspects of Neural Networks, with links to statistical mechanics and optimization methods; for example, I have worked out a theoretical framework to unify Energy based and Feed Forward model. The framework is used to obtain the «natural» activation functions that maximize the information flow in the Neural Network and therefore lead to improved optimization protocols. Currently, I am analytically investigating the Complexity of Feed Forward networks and the geometry of their loss manifold using tools developed to study Spin Glass physics.

Before moving into AI, I worked in condensed matter theory with a focus on the interplay between strong interactions, topology and non-equilibrium. Here I used a varieties of tools such as Quantum Field theory to investigate the behavior of quantum matter in strongly interacting and disordered systems.