Statistical Physics in Inference and (Deep) Learning

In this course students will explore the statistical physics principles underlying probabilistic inference and various neural network architectures. The course is designed to bridge the gap between teaching approaches to modern statistical physics that are either purely theoretical, or focus largely on its applications in data analysis. To that end, there will be a conscious effort to study topics such as: MaxEnt principle, variational methods, Hebb’s rule, bias-variance tradeoff, regularization, and others with analytical derivations as well as worked-out code examples in Jupyter notebooks. The course will also provide a space for students to interrogate and reflect on the ethical, political, and policy frameworks that are urgently needed in the age of deep learning.