A0449
Title: Learning of deep convolutional network image classifiers via stochastic gradient descent and over-parametrization
Authors: Michael Kohler - Technische Universitaet Darmstadt (Germany)
Adam Krzyzak - Concordia University (Canada)
Alisha Saenger - Technische Universität Darmstadt (Germany) [presenting]
Abstract: Image classification from independent and identically distributed random variables is considered. Image classifiers are defined based on a linear combination of deep convolutional networks with a max-pooling layer. All the weights are learned by stochastic gradient descent. A general result is presented, which shows that the image classifiers are able to approximate the best possible deep convolutional network. In case the a posteriori probability satisfies a suitable hierarchical composition model, it is shown that the corresponding deep convolutional neural network image classifier achieves a rate of convergence independent of the dimension of the images.
