Lars Ruthotto

Short Course on Numerical Methods for Deep Learning

Description:
This short course gives an introduction into numerical methods for training deep neural networks. The course consists of three lectures. The first lecture introduces the learning problem and considers linear models. The second lecture discusses traditional neural networks and in particular residual neural networks. The last lecture, uses the similarity between residual neural networks and nonlinear time-dependent to motivate optimal control methods for their training.
This course is presented to a group of PhD students in the Summer School in Chemnitz.
Acknowledgement:
Thanks to the organizers for the kind invitation and covering my travel expenses.
Times:
September 21-22, 2018
Slides:
0-Notation.pdf
0-Literature.pdf
1-LinearModels.pdf
2-NonlinearModels.pdf
3-ContinuousNeuralNetworks.pdf
Codes:
The examples from class can be reproduced and extended using the files in NumDL MATLAB code.