Title:	Custom Machine learning Algorithm benchmarking
DNr:	SNIC 2018/7-60
Project Type:	SNIC Small Compute
Principal Investigator:	Antoine Honoré <honore@kth.se>
Affiliation:	Karolinska Institutet
Duration:	2018-10-22 – 2019-11-01
Classification:	10201
Keywords:

Abstract

Here we try to unify classical deep learning and kernel methods. Kernel methods are known to perform well when the amount of training data is limited. On the other hand Deep learning is able to handle a large amount of training data to optimize deep network weights. We designed a random weights based network which requires only one regularized parameter to be learnt at each layer. Our approach is mathematically founded and the developed network is expected to show good generalization power due to appropriate regularization and use of random weights in the layers.