Symposium on Learning, Mathematics and Physics
This semester the Amsterdam Math-Physics Colloquium will be expanded to a symposium on Learning, Mathematics and Physics. The sympostium is organised by Miranda Cheng, co-organised with Prof. Max Welling and his team members Roberto Bondesan, Pim de Haan, and Daniel Worall.
Event details from: Symposium on Learning, Mathematics and Physics
- Date
- 8 November 2019
- Time
- 10:30 – 16:30
Programme
10.30 - 11.00 Coffee/Tea Break
11.00 - 12.00 Danilo Rezende (DeepMind) : Generative Models and Symmetries
12.00 - 13.45 Lunch
13.45 - 14.15 Coffee/Tea Break
14.15 - 15.00 Clément Hongler (EPFL Lausanne): Neural Tangent Kernel and Applications
15.00 - 15.30 Coffee/Tea Break
15.30 - 16.30 Stéphane Mallat (Collège de France, ENS Paris): Multiscale Interactions in Physics and Machine Learning with Deep Nets
Abstracts
Danilo Rezende (DeepMind) : Generative Models and Symmetries
In this talk I will discuss how some ideas from Physics such as phase-transitions and Gauge symmetries provide powerful tools to analyse and build generative models. In particular, the study of symmetries in Physics has revolutionised our understanding of the world and permeates every fundamental physical model. Inspired by this, I will focus on our recent work of incorporating Gauge symmetries into normalizing flows generative models and its potential applications in the sciences and ML.
Clément Hongler (EPFL Lausanne): Neural Tangent Kernel and Applications
The Neural Tangent Kernel is a new way to understand the gradient descent in deep neural networks, connecting them with kernel methods. In this talk, I'll introduce this formalism and give a number of results on the neural tangent kernel and explain how they give us insight into the dynamics of neural networks during training and into their generalization features. Based off joint works with Arthur Jacot and Franck Gabriel.
Stéphane Mallat (Collège de France, ENS Paris): Multiscale Interactions in Physics and Machine Learning with Deep Nets
Is there a generic notion of regularity allowing to generate low-dimensional models for high-dimensional problems in physics and machine learning ? There is no mathematical answer to this outstanding question but the success of deep neural networks architectures indicates that such a regularity exists. This talk outlines the role of scale separation, symmetries and sparsity, to define models implemented by convolutional neural networks. I will discuss open problems to regress quantum molecular energies and to define statistical physics models of turbulence, face to face with image classification and generation problems, which appear to have similar regularities.
Location
CWI, Eulerzaal