Description Together with Zack Xuereb Conti, I co-founded and co-organize an online seminar series hosted by The Alan Turing Institute, which began in 2021. The seminar series runs every other week, excluding a summer break, with the flexibility to accommodate additional dates. Currently, we have over 1,000 subscribers and an average attendance of 80 people. The talks are uploaded on The Alan Turing Institute's official YouTube channel.
Description The aim of this workshop is to present the results obtained during the relevant scientific retreat, held online and hosted by The Alan Turing Institute in October 2023.
Topic This online workshop delves into the application of Statistical Mechanics (SM) to Artificial Neural Networks (ANN) and Explainable AI (XAI). We explore how SM principles, pivotal in Theoretical Physics, apply to neural networks, emphasizing collective phenomena in both biological and artificial systems. Key topics include the role of Restricted Boltzmann Machines (RBM) in learning architectures, the Hebbian learning rule, and its applications in machine learning. Our aim is to understand emergent properties in deep networks, their spectral properties to avoid overfitting, and the potential integration of feed-forward networks within SM frameworks. This workshop promises to advance our theoretical understanding of AI through the lens of Statistical Mechanics.
Description Together with Adriano Barra and Elena Agliari, I co-organized and served as the main coordinator for a two-week in-person scientific retreat hosted at The Alan Turing Institute in January 2022. This retreat brought together researchers, postdocs, and PhD students. Topic The scientific retreat focused on the topic of Statistical Mechanics (SM), which provides a probabilistic formulation of the macroscopic behavior of systems composed of many interacting microscopic entities. This discipline, a cornerstone of Theoretical Physics, emerged in the past century to characterize collective phenomena such as phase transitions. Notably, features of biological neural networks (like memory, computation, and other emergent skills) can be understood within the framework of SM once the mathematical modeling of its elemental constituents (neurons, axons, synapses, etc.) is established. Individual neurons cannot recognize information patterns; this ability emerges from their interactions. Since the pioneering work of Amit, Gutfreund, and Sompolinsky, SM of Disordered and Complex Systems has been crucial in understanding information processing in Artificial Neural Networks (ANN). It is also expected to play a vital role in the development of Explainable Artificial Intelligence (XAI), particularly in the new generation of (potentially "deep") neural networks and learning machines. This workshop maintained an SM perspective, integrating mathematical and theoretical physics with machine learning.