Effect of Batch Normalization on Noise Resistant Property of Deep Learning Models

Abstract

The fast execution speed and energy efficiency of analog hardware has made them a strong contender for deployment of deep learning model at the edge. However, there are concerns about the presence of analog noise which causes changes to the weight of the models, leading to performance degradation of deep learning model, despite their inherent noise resistant characteristics. The effect of the popular batch normalization layer on the noise resistant ability of deep learning model is investigated in this work. This systematic study has been carried out by first training different models with and without batch normalization layer on CIFAR10 and CIFAR100 dataset. The weights of the resulting models are then injected with analog noise and the performance of the models on the test dataset is obtained and compared. The results show that the presence of batch normalization layer negatively impacts noise resistant property of deep learning model and the impact grows with the increase of the number of batch normalization layers.

The method of building detection in remote sensing images based on improved Segformer

“…1) The detection network structure is Optimized, and the construction of the network model aiming at the building features is intensified. Multiple normalization activation layers [9] are fused, and the over-fitting of the network is averted by compressing the number of parameters and restraining gradient disappearance. Meanwhile, effective screening and classification of the feature information within the classification can be realized by embedding the activation layer, to more effectively gain the semantic information features for buildings.…”