Dr Mattia Salvardi delivered a seminar entitled. « From Energy Landscapes to Learning Machines: A Physics-Inspired Introduction to Neural Networks and Energy-Based Models » on Monday 12 May 2025 & Wednesday, 14 May 2025.

Abstract. This seminar explores the fundamental principles of neural networks through the lens of energy minimization, drawing inspiration from classical physics. We begin with an intuitive introduction to neural networks, highlighting their evolution from fixed-feature models to flexible, learning-driven architectures. Through visualizations and mathematical derivations, we examine the role of gradient descent, cost landscapes, and backpropagation in training dynamics. The seminar then extends to energy-based models (EBMs), tracing their development from Hopfield networks to modern deep energy-based approaches. Finally, an interactive session with real-time simulations will provide hands-on insights into learning dynamics and hyperparameter tuning.

Bio. Dr. Mattia Savardi is a Research Fellow and Associate Lecturer at the Department of Medical and Surgical Specialties, Radiological Sciences, and Public Health, University of Brescia. He received his M.Sc. in Communication Technologies and Multimedia (cum laude) and obtained his Ph.D. with merit in Technology for Health at UNIBS, focusing on Deep Learning for Medical Image Analysis. Awarded the GTTI PhD prize in 2020 for the best thesis, he has worked with brain functional MRI, CXR, hyperspectral, and RGB biomedical images in collaboration with international institutions. His research spans media psychology—analyzing formal cinematic features that elicit measurable effects on viewers—and neuroscience, where he explored brain decoding through fMRI to identify film features perceived during viewing sessions.