Over the past two decades, deep learning and neural networks have been increasingly successful in machine learning tasks.

Despite these successes, the principles underlying deep learning remain a formidable challenge for theorists. The reasons for their surprising generalisability, their optimal architecture and parameterisation, and a classification of different learning tasks based on their tractability are still largely open problems. 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.

This second edition of the school will primarily address the large audience of early-stage researchers (graduate students, advanced master students, and postdocs) in physics and applied mathematics who are interested in fundamental aspects of deep learning and computational neuroscience 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 participants to a number of recent applications to a range of problems, mainly of a physical nature.