Learning Physics with Deep Neural Networks

Can we learn physical properties from data? Machine learning offers a solution. It has many similarities with physics, requiring the approximation of functionals which depend on large numbers of variables, such as millions of pixels in images, letters in text, or particles in a physical system. Machine learning algorithms have considerably improved in the last 10 years through the processing of massive amounts of data. In particular, deep neural networks have spectacular applications, such as image classification and medical, industrial and physical data analysis.

In this lecture, Stéphane Mallat will show how machine learning can be applied to statistical physics, turbulent fluids and quantum chemistry. Beyond applications, he will highlight common mathematical approaches in physics and machine learning to overcome the issue of dimensionality. Two central pillars of such approaches are finding symmetries and separating phenomena at different scales. He will show that these pillars also govern the architecture and properties of deep convolutional neural networks.