A Two-Stage Subspace Trust Region Approach for Deep Neural Network Training

no code implementations • 23 May 2018 • Viacheslav Dudar, Giovanni Chierchia, Emilie Chouzenoux, Jean-Christophe Pesquet, Vladimir Semenov

In this paper, we develop a novel second-order method for training feed-forward neural nets.

Use of symmetric kernels for convolutional neural networks

no code implementations • 23 May 2018 • Viacheslav Dudar, Vladimir Semenov

At this work we introduce horizontally symmetric convolutional kernels for CNNs which make the network output invariant to horizontal flips of the image.