Pattern-based convolutional neural network pruning method and pattern sensing accelerator

Abstract

The invention discloses a pattern-based convolutional neural network pruning method and a pattern sensing accelerator. Firstly, a pattern-based convolutional neural network (PCNN) filter is provided,regular pruning in the filter is realized based on a multi-knapsack framework, and the method can be orthogonal to an existing pruning method, is similar to fine-grained sparsity, but still keeps certain regularity to a certain extent. In each layer, the number of non-zeros in the filter is limited to be the same, so that the calculation workloads of different convolution windows can be the same.Based on the uniform sparsity ratio in the filter, the pattern type in one layer is very limited. The filters are described with pattern masks with very few bits and corresponding non-zero sequences rather than all values, thereby significantly reducing memory size and computational complexity, and being very friendly to both accuracy and hardware. The PCNN pruning method can be orthogonal to other pruning methods such as filter pruning, the PCNN pruning method can realize nine-time weight pruning, and the accuracy loss can be ignored.

Description

technical field [0001] The invention relates to a convolutional neural network method and hardware optimization, in particular to a pattern-based convolutional neural network pruning method and a pattern-aware accelerator. Background technique [0002] Neural network is one of the most important algorithms in the field of artificial intelligence at present, and it can be widely used in fields such as face recognition, self-driving and health care. Traditionally, a neural network consists of three distinct layers, including a convolutional layer, a pooling layer, and a fully connected layer, which implement feature extraction, downsampling, and classification functions, respectively. With the development of network architectures, convolutional layers have become the main bottleneck of computation and storage. Therefore, in order to bring artificial intelligence to various fields with lower overhead, it is urgent to solve a large number of calculations and parameters. [000...

AI Technical Summary

Problems solved by technology

This approach is not sufficient to maintain the accuracy of the entire model, since error propagation through the layers is ignored

Furthermore, experiments are limited to small neural networks and datasets, such that the resulting architectures cannot skip redundant computations and thus cannot accelerate computations

Images