A Novel Deep-Learning Model Compression Based on Filter-Stripe Group Pruning and Its IoT Application

Abstract

Nowadays, there is a tradeoff between the deep-learning module-compression ratio and the module accuracy. In this paper, a strategy for refining the pruning quantification and weights based on neural network filters is proposed. Firstly, filters in the neural network were refined into strip-like filter strips. Then, the evaluation of the filter strips was used to refine the partial importance of the filter, cut off the unimportant filter strips and reorganize the remaining filter strips. Finally, the training of the neural network after recombination was quantified to further compress the computational amount of the neural network. The results show that the method can significantly reduce the computational effort of the neural network and compress the number of parameters in the model. Based on experimental results on ResNet56, this method can reduce the number of parameters to 1/4 and the amount of calculation to 1/5, and the loss of model accuracy is only 0.01. On VGG16, the number of parameters is reduced to 1/14, the amount of calculation is reduced to 1/3, and the accuracy loss is 0.5%.

Empirical Compression Features of Mobile Computing and Data Applications Using Deep Neural Networks

“…But the hierarchical model does not provide a structural view with any analytical framework; thus, a separate algorithm is integrated. Even in recent times, the researchers have directed the research model using a deep neural network where the design is completely based on complementary mobile terminals [13][14][15][16][17]. If each terminal in the mobile nodes is stationary, then insignifcant data flters will be removed from the system.…”

UNet++ Compression Techniques for Kidney and Cyst Segmentation in Autosomal Dominant Polycystic Kidney Disease

“…Pruning is one of the widely used compression techniques, which eliminates redundant connections or filters from the model to reduce its size without compromising performance [6]. However, sparse connectivity in the network resulting from pruning can lead to irregular memory access patterns and reduced performance during inference [10]. Implementing sparse networks on hardware, particularly on embedded devices with limited memory resources, can also present challenges.…”