Statistical Physics of
Deep Learning III
15-19 June 2026

School description

Over the last two decades, deep learning and neural networks enjoyed a growing success, ranging from classification tasks (such as image recognition) to generative models at the forefront of the current AI revolution (large language models) and applications in computational neurosciences. In the context of physical sciences, deep learning algorithms have been applied to a wide array of different problems faced by physicists, from particle physics and cosmology to many-body physics and biological physics. It is thus becoming an increasingly important aspect in the training of many physicists and applied mathematicians.

Despite these empirical successes, the principles underlying deep learning remain largely elusive. Typical deep neural networks used in applications function as black boxes, making it difficult to understand theoretically the rationale for their impressive performances and to interpret how they reach specific decisions. The reason of their surprising generalization capability, their optimal architecture and parametrization and a classification of different learning tasks based on their tractability are still largely open problems. From a computational neuroscience perspective, the theoretical study of information processing in deep networks paves the way for a better understanding of the mechanisms of learning and memory in biological brains. Among other approaches, methods rooted in statistical physics, such as disordered systems, phase transitions or chaos theory, have begun to provide conceptual insights into these questions.

Following the successful editions of 2022 and 2024, we will host the third edition of our school on the statistical physics of deep learning on June 15-19 2026.

The school will be aimed primarily at the large audience of early-stage researchers (graduate students, advanced master students and postdocs) in physics and applied mathematics interested in fundamental aspects of deep learning and computational neurosciences, beyond a simple black-box approach. Extended lectures, tutorials, and research seminars will provide a critical introduction to these topics from a statistical physics perspective and will expose the participants to a number of current research problems at the forefront of the field.

 

A beautiful location

Lake Como School of Advanced studies is located c/o Fondazione Alessandro Volta in the beautiful setting of Villa del Grumello, in Como, Italy

Venue & Accommodation

The Lake Como School of Advanced Studies is an international research facility. We run fellowships, short term programmes on a wide range of interdisciplinary subjects, that share a common focus on complex systems.