Unlock instant, AI-driven research and patent intelligence for your innovation.
Convolutional neural network operation system, method and device
What is Al technical title?
Al technical title is built by PatSnap Al team. It summarizes the technical point description of the patent document.
A convolutional neural network and computing system technology, applied in neural learning methods, biological neural network models, neural architectures, etc., can solve problems such as model storage, long time consumption of all models, and unobvious reduction of hardware resources, so as to improve computing ability, hardware cost saving effect
Pending Publication Date: 2021-05-11
GREE ELECTRIC APPLIANCES INC +1
View PDF0 Cites 0 Cited by
Summary
Abstract
Description
Claims
Application Information
AI Technical Summary
This helps you quickly interpret patents by identifying the three key elements:
Problems solved by technology
Method used
Benefits of technology
Problems solved by technology
However, commonly used convolution model weight data such as figure 2 As shown, there are usually hundreds of megabytes, such as figure 2 The storage resources of the shown AleNet, VGG16 and Inception-v3 network models all exceed 100 megabytes, so it is difficult to store the entire model in a small device, and a large internal storage unit must be used to store the model completely, resulting in hardware costs for convolution operations High, and because the model weight data is large, it takes a long time to read all the models, which affects the computing power of the computing unit
[0004] Compressing the convolution model is a new direction to save hardware resources, but the overall compression and overall decompression methods are commonly used in the existing technology, and the reduction of hardware resources is not obvious. Therefore, a compression method is urgently needed to significantly reduce hardware resources and save hardware cost
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 more
Image
Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
Click on the blue label to locate the original text in one second.
Reading with bidirectional positioning of images and text.
Smart Image
Examples
Experimental program
Comparison scheme
Effect test
Embodiment 1
[0060] Such as Figure 5 As shown, it is a schematic flow diagram of using the Huffman compression algorithm. Huffman coding is an entropy coding technology proposed by DavidA.Huffman, and its coding method is as follows:
[0061] Set a binary code for each symbol in the signal source, the symbols with higher frequency will get shorter bits, and the symbols with lower frequency will be allocated longer bits, so as to improve the compression rate and reduce hardware storage resources .
[0062] The specific encoding method is as follows:
[0063] S501, initialize the model matrix.
[0064] S502, obtain the current address, and judge whether to enter the current array, if it is determined to enter, proceed to S503, otherwise proceed to S504, and enter the next address.
[0065] S503, enter the current array, and execute step S505.
[0066] S504, enter the next address, and execute step S502.
[0067] S505, judging whether the entry address is over, if it is over, go to step...
Embodiment 2
[0071] The specific encoding method of the LZ77 compression algorithm is as follows:
[0072] Initialize the coding position first, judge whether the current coding position is the last model address, if the current coding position is the last model address, then complete, otherwise perform LZ77 coding; after starting LZ77 coding, first judge the character, if the characters are the same, then the model address Increment by 1, otherwise enter the last character judgment, if it is true with the last character, then the model address will increase the length of a specific number, if the position of the nth and n+1th times are equal (C[N]=C [N+1]), then the current encoding is merged, and the address accumulation LEN is performed in the next step.
[0073] S402. Based on the pre-set operation sequence and convolutional neural network model resources, sequentially determine the convolutional layer model used for the convolutional neural network operation of each layer.
[0074] D...
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 discloses a convolutional neural network operation system, method and device, which are used for saving hardware cost of convolutional neural network operation and improving operation capability, and the system comprises a model decoding module which is used for decoding the externally transmitted convolutional neural network model resources to obtain a convolutional layer model, wherein the convolutional layer model is stored in a model storage module; a model storage module used for storing a convolutional layer model; and a data storage module used for storing the image data and the operation result data, performing convolutional neural network operation on the image data stored in the data storage module based on a convolutional layer model, and storing the operation result data after operation in the data storage module.
Description
technical field [0001] The present invention relates to the technical field of deep learning, in particular to a convolutional neural network computingsystem, method and equipment. Background technique [0002] Convolutional Neural Network (CNN), as a kind of artificial neural network, is a kind of feed-forward neural network that includes convolution calculation and has a deep structure. It is one of the representative algorithms of deep learning. Convolutional neural network has Representation learning ability, which can perform translation-invariant classification of input information according to its hierarchical structure, has become a research hotspot in the field of speech analysis and image recognition. [0003] Such as figure 1 As shown, in the prior art, the convolutional neural network operation system 10 usually needs to store the trained model in the hardware before performing the convolution operation. During the convolution operation, all model parameters ar...
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
Application Date:The date an application was filed.
Publication Date:The date a patent or application was officially published.
First Publication Date:The earliest publication date of a patent with the same application number.
Issue Date:Publication date of the patent grant document.
PCT Entry Date:The Entry date of PCT National Phase.
Estimated Expiry Date:The statutory expiry date of a patent right according to the Patent Law, and it is the longest term of protection that the patent right can achieve without the termination of the patent right due to other reasons(Term extension factor has been taken into account ).
Invalid Date:Actual expiry date is based on effective date or publication date of legal transaction data of invalid patent.
Login to View More 
Login to View More