Neural network-oriented matrix conversion device and method
A matrix conversion and neural network technology, applied in the computing field, can solve problems such as increasing the difficulty of neural network computing, excessive repetitive operation tasks, wasting computing resources, etc., and is conducive to widespread use, improving computing speed, and simplifying computing methods Effect
Active Publication Date: 2021-07-20
INST OF COMPUTING TECH CHINESE ACAD OF SCI
View PDF3 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0003] The calculation process of the neural network generally includes steps such as convolution, activation, and pooling. According to research, when performing the matrix operation of the convolution process, the repetitive operation tasks for the data are extremely large, especially the multiplication operation of the matrix. Not only increases the difficulty of neural network calculations, but also wastes a lot of computing resources due to repeated calculations, resulting in a decrease in computing speed
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
[0042] In order to make the objects, technical solutions, and advantages of the present invention, a matrix conversion apparatus and method provided in the embodiment of the present invention will be described in further detail below with reference to the accompanying drawings.
[0043] Since the convolutionary operation in the neural network is more operation, the number of convolutional operations is proportional to the number of movements moved by the convolutionary window. After research, scholars have proposed a Winograph-based convolutionary computation method, which uses a specific conversion matrix to perform matrix conversion of the input characteristic image data with the weight data, thereby completing the equivalent convolutional calculation task for Reduce a large number of multiplication processes during the convolutionary operation, so how to design an efficient calculation device for matrix conversion has become a research focus.
[0044] In general, Winograd-based...
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 relates to a neural network-oriented matrix conversion device, comprising: a data receiving interface for receiving and transmitting the matrix data to be converted of the neural network; a conversion matrix receiving interface for receiving and transmitting the data corresponding to the matrix to be converted The transformation matrix; matrix operation unit, is respectively connected with described data receiving interface and described transformation matrix reception interface, utilizes described to-be-transformed matrix data and described corresponding transformation matrix to carry out matrix transformation operation; Output interface, is connected with described matrix The operation unit is connected to splice and output the operation results obtained by the matrix operation unit; the temporary storage unit is connected to the data receiving interface and the output interface, and is used to temporarily store the operation results output by the output interface, and Input the operation result to the data receiving interface.
Description
Technical field [0001] The present invention relates to the field of calculations, and more particularly to a matrix conversion device and method for neural networks. Background technique [0002] Neural networks are one of the high development levels in the artificial intelligence. Because of a wide range of applications and excellent performance, it has become a research hotspot in academic and industrial communities. The neural network creates a model structure by simulating the neural connection structure of the human brain, bringing breakthrough progress for large-scale data (such as images, video or audio) processing tasks, which is a network of networks through a large number of nodes. The arithmetic model is configured, and the node is referred to as a neuron. The connection strength between each two nodes represents the weighting value between the two nodes in the two nodes, corresponding to the human neural network. memory. [0003] The calculation process of neural net...
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 Patents(China)
IPC IPC(8): G06N3/04G06F17/16
CPCG06F17/16G06N3/045
Inventor 韩银和闵丰许浩博王颖
Owner INST OF COMPUTING TECH CHINESE ACAD OF SCI