1. A Robust Neural Network with Simple Architecture.
- Author
-
Timofeev, V. S. and Sivak, M. A.
- Abstract
Under consideration are the classification problem and application of simple neural networks for solving the problem. A robust modification of the error backpropagation algorithm is proposed and used for training neural networks. Some proposition is proved that allows us to construct the proposed modification with the Huber loss-function. In order to study the properties of the so-obtained neural network, a number of computational experiments are carried out. We consider various values of the outliers' fraction, noise level, and training and test samples sizes. Inspection of the results shows that the proposed modification can significantly increase the classification accuracy and learning rate of a neural network when working with noisy data. [ABSTRACT FROM AUTHOR]
- Published
- 2021
- Full Text
- View/download PDF