Training the Convolutional Neural Network with Statistical Dependence of the Response on the Input Data Distortion

TL;DR

This paper introduces a novel training method for CNNs that incorporates input data distortion levels, enhancing robustness without altering the original network architecture, demonstrated on MNIST with Gaussian blur.

Contribution

The approach adds a temporary layer during training to encode distortion information, improving robustness while maintaining the original network structure.

Findings

Significant reduction in error rates on distorted test data.

Presence of a strong statistical dependence between responses and input distortions.

No quality loss in network performance on original data.

Abstract

The paper proposes an approach to training a convolutional neural network using information on the level of distortion of input data. The learning process is modified with an additional layer, which is subsequently deleted, so the architecture of the original network does not change. As an example, the LeNet5 architecture network with training data based on the MNIST symbols and a distortion model as Gaussian blur with a variable level of distortion is considered. This approach does not have quality loss of the network and has a significant error-free zone in responses on the test data which is absent in the traditional approach to training. The responses are statistically dependent on the level of input image's distortions and there is a presence of a strong relationship between them.