Statistical Physics in Machine Learning: Understanding Deep Learning

This lecture by the instructor delves into the application of statistical physics concepts in understanding deep learning, focusing on neural networks. Starting with the theory of learning with neural networks, the lecture covers the working principles of neural networks, hierarchy of features, and the core of machine learning today. It explores the sample complexity, principal learning theory results, and the challenges in understanding deep learning. The lecture also discusses the teacher-student model, learning curves, phase transitions, and the quest towards a theory of deep learning. Emphasizing the importance of physics in deciphering deep learning, the lecture concludes by highlighting the potential of physics tools in data science and the mutual benefits between physics and related developments.