MO3.R2.4

Rate of convergence of an over-parametrized convolutional neural network image classifier learned by gradient descent

Michael Kohler, Technical University of Darmstadt, Germany; Adam Krzyzak, Concordia University, Canada; Benjamin Walter, Technical University of Darmstadt, Germany
Session:
Classification and Regression
Track:
8: Machine Learning
Location:
Ypsilon I-II-III
Presentation Time:
Mon, 8 Jul, 15:35 - 15:55
Session Chair:
Adam Krzyzak, Concordia University
Abstract
Image classifiers based on over-parametrized deep convolutional neural networks with an average-pooling are proposed. The weights of the network are learned by gradient descent. We present the bound on the rate of convergence of the difference between the expected misclassification risk of the plug-in classifier and the Bayes risk. The obtained rate of convergence is independent of image dimension under appropriate constraints on the image distribution.
Session MO3.R2
MO3.R2.1: Effect of Weight Quantization on Learning Models by Typical Case Analysis
Shuhei Kashiwamura, The University of Tokyo, Japan; Ayaka Sakata, The Institute of Statistical Mathematics, Japan; Masaaki Imaizumi, The University of Tokyo, Japan
MO3.R2.2: Sharp information-theoretic thresholds for shuffled linear regression
Leon Lufkin, Yihong Wu, Yale University, United States; Jiaming Xu, Duke University, United States
MO3.R2.3: Data-Driven Estimation of the False Positive Rate of the Bayes Binary Classifier via Soft Labels
Minoh Jeong, Martina Cardone, University of Minnesota, United States; Alex Dytso, Qualcomm Flarion Technology, Inc., United States
MO3.R2.4: Rate of convergence of an over-parametrized convolutional neural network image classifier learned by gradient descent
Michael Kohler, Technical University of Darmstadt, Germany; Adam Krzyzak, Concordia University, Canada; Benjamin Walter, Technical University of Darmstadt, Germany
Resources
View Manuscript