Statistical Physics of Computation Laboratory

At the Statistical Physics of Computation Laboratory we view computational problems and associated algorithms as systems of many interacting elements (e.g. bits, weights, components of a signal, neurons). We use tools of statistical physics and mathematics to describe and understand their behaviour, performance and limitations. We then use the obtained insights to develop better algorithms. We are interested in computational problems in a broad range of areas, in physics itself, in machine learning with deep neural networks, statistical inference, optimization, combinatorics, epidemic spreading, signal processing, etc. Currently, we focus on statistical physics of learning, developing solvable models and theoretical principles that explain how modern AI systems generalize, memorize, and scale.

A short video about our research. A video by the students about our group.