Training the Convolutional Neural Network with Statistical Dependence of the Response on the Input Data Distortion
This work addresses robustness to input distortions for image classification tasks, but it is incremental as it builds on existing architectures like LeNet5 and focuses on a specific distortion model.
The paper tackled the problem of training convolutional neural networks to handle distorted input data by modifying the learning process with an additional layer that is later removed, preserving the original architecture. The result showed no quality loss and a significant error-free zone in test responses, with strong statistical dependence between responses and distortion levels.
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.