A method for accelerating convolutional neural network
A convolutional neural network and convolution kernel technology, applied in biological neural network models, neural architectures, instruments, etc., can solve problems such as unbalanced resource allocation, reduce the overall computing efficiency of the convolution layer, and reduce the amount of computation and complexity. degree, increase universality, and balance the amount of computation
Pending Publication Date: 2020-07-28
INST OF COMPUTING TECH CHINESE ACAD OF SCI
View PDF0 Cites 2 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
However, in depth-separable convolution, since the algorithm complexity of point-by-point convolution is much higher than that of task depth convolution, a large amount of resources are used for information interaction between different feature layers, resulting in unbalanced resource allocation, thus Reduced the overall operational efficiency of the convolutional layer
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0026] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.
[0027] The depthwise separable convolution in the prior art decomposes a complete convolution operation into two steps of depthwise convolution and point-by-point convolution, wherein the depthwise convolution is performed by using a single The channel convolution kernel performs filtering, that is, one convolution kernel is only responsible for one channel. Figure 1A A schematic diagram of the depthwise convolution operation in depthwise separable convolution is shown. Such as Figure 1A As shown, assume that the size of the input feature map is W*H, the number of channels (layers) is N, the size of the convolution kernel is K*K, and the number of convolution kernels is equal to the number of channels of the input feature map. When the step size is 1 and the padding is 0, an output feature map with a size of W*H and a channel number of N ca...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention provides a method for accelerating a convolutional neural network. The method comprises the steps that 1, an input feature map with N channels is divided into G groups of initial featuremaps in the channel direction, the Gith group of initial feature maps comprise Si feature maps, first group convolution is conducted on the G groups of initial feature maps to obtain G groups of first feature maps, and N, G and Si are integers larger than or equal to 1; 2, the G groups of first feature maps are re-divided into F groups of second feature maps, wherein the Fjth group of second feature maps comprises Tj feature maps from different first feature map groups, second group convolution is carried out on the F groups of second feature maps to obtain an output feature map with M channels, and F, Tj and M are integers greater than or equal to 1.
Description
technical field [0001] The invention relates to deep learning technology, in particular to a method for accelerating convolutional neural networks. Background technique [0002] With the development of deep learning technology, the application technology based on deep learning is widely used in various fields of life. Deep learning has also gradually developed from the earliest cloud computing to today's terminal computing. Due to the large scale of most deep learning applications, high performance requirements for machines during training and prediction, and limited storage resources and computing power of terminal devices, how to accelerate deep learning has become a new technology hotspot. Convolutional neural network is a commonly used deep learning model, and its acceleration method is one of the hot topics. [0003] Convolutional Neural Network (CNN) is similar to the multi-layer perceptron of artificial neural network, which extracts features through convolution, in...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/46G06K9/00G06N3/04
CPCG06V10/94G06V10/454G06N3/045
Inventor 陈尧麟郝昀超张佩珩霍志刚
Owner INST OF COMPUTING TECH CHINESE ACAD OF SCI