https://zoom.us/j/475819702

Meeting ID: 475-819-702

If you haven't registered for previous QLS webinars, please contact qls@ictp.it to obtain the PASSWORD for this zoom meeting.
 
Abstract: Understanding the impact of data structure on learning in neural networks remains a key challenge for the theory of neural networks.
In these two lectures, we will discuss how to go beyond the simple i.i.d. modelling paradigm in the teacher-student setup by studying neural networks trained on data drawn from structured generative models. 

Our discussion will center around two results:
(1) We give rigorous conditions under which a class of generative models shares key statistical properties with an appropriately chosen Gaussian feature model.
(2) We use this Gaussian equivalence to analyse the dynamics of two-layer neural networks trained using one-pass stochastic gradient descent on data drawn from a large class of generators.   

I will try to make these lectures self-contained. 

They will be mostly based on the following two papers: 
[1] Goldt, S., Mézard, M., Krzakala, F. and Zdeborová, L., 2020.
     Modeling the Influence of Data Structure on Learning in Neural Networks: The Hidden Manifold Model. Physical Review X, 10(4), p.041044.
     https://arxiv.org/abs/1909.11500
[2] Goldt, S., Reeves, G., Loureiro, B., Mézard, M., Krzakala, F. and Zdeborová, L., 2020.
     The Gaussian equivalence of generative models for learning with two-layer neural networks. 
     under review; https://arXiv.org/abs/2006.14709