Quantitative calculation method and system for convolutional neural network
A convolutional neural network and calculation method technology, applied in the field of neural network algorithm hardware implementation, can solve problems such as low accuracy, large array power consumption, and insufficient computing power, so as to improve speed, reduce calculation power consumption, and increase The effect of throughput
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Example Embodiment
[0084] Example 1
[0085] Such as Figure 7 In this embodiment, the AlexNet network is taken as an example to further illustrate the quantization calculation method of the convolutional neural network of the present invention:
[0086] The AlexNet network structure has a total of 8 layers, the first 5 layers are convolutional layers, and the next three layers are fully connected layers; the convolutional neural network quantitative calculation method accelerates the network into three steps:
[0087] The first step is to perform high-precision quantization on the first layer of convolution (Conv1), the second layer of convolution (conv2), the penultimate fully connected layer (Fc7), and the last fully connected layer (Fc8). Layer convolution (Conv3), fourth layer convolution (Conv4), fifth layer convolution (Conv5), 6th layer full connection (Fc6) for binarization, whether the accuracy of the software simulation simulation operation meets the requirements, For example, whether the a...
Example Embodiment
[0093] Example 2
[0094] Such as Picture 8 This embodiment takes the LeNet network as an example to further illustrate the quantization calculation method of the convolutional neural network of the present invention: the LeNet network is simple and only has 7 layers in total, including the convolutional layer (Conv), the pooling layer (pool) and the fully connected layer (Fc ), the convolutional neural network quantitative calculation method to accelerate the network includes three steps:
[0095] In the first step, the computationally intensive layers (convolutional layer, fully connected layer) are divided into binary quantization and high-precision quantization. Generally, the first layer of convolution (Conv1) and the last layer are fully The connection layer (Fc2) performs high-precision quantification, and the second layer convolution (Conv2) and the first layer fully connected layer (Fc1) are binarized and quantified. After the software simulation runs, whether the accurac...
Example Embodiment
[0101] Example 3
[0102] Such as Picture 9 In this embodiment, the DeepID1 network is taken as an example to further illustrate the quantization calculation method of the convolutional neural network of the present invention:
[0103] The DeepID1 neural network model used to extract facial features in face recognition algorithms is mainly composed of convolutional layer (Conv), pooling layer (pool) and fully connected layer (Fc). Convolutional neural network quantitative calculation method is accelerated The network consists of three steps:
[0104] In the first step, the computationally intensive layers (convolutional layer, fully connected layer) are divided into binary quantization and high-precision quantization. Generally, the first layer of convolution (Conv1) and the last layer are fully The connection layer (Fc) performs high-precision quantization, and the second layer of convolution (Conv2), the third layer of convolution (Conv3), and the fourth layer of convolution (Con...
PUM
Abstract
Description
Claims
Application Information
- R&D Engineer
- R&D Manager
- IP Professional
- Industry Leading Data Capabilities
- Powerful AI technology
- Patent DNA Extraction
Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.
© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap