Multi-parallel acceleration method for CNN full connection layer operation
A fully connected layer and computing result technology, applied in neural architecture, biological neural network models, etc., can solve problems such as reducing accuracy, wasting energy consumption, consuming processing time, and reducing the number of parameters, so as to improve the system energy efficiency ratio and save Effects of processing time and guaranteed processing accuracy
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment
[0048] The fully-connected layer operation acceleration method based on programmable logic circuits designed by the present invention is divided into the following modules: a processing data format determination module, an operation parallel degree determination module, a fully-connected layer weight parameter data format conversion and storage module, and an operation flow control module , a data judgment module, a data storage control module A / B, a data calculation module, a calculation result storage module and a data output module.
[0049] 1. The processing data format determination module is used to determine the processing data format: according to the application scenario and target characteristics of CNN, determine the data format used for the operation of the fully connected layer. If CNN is used to classify and identify remote sensing ship targets, the classification types are designed to be at most 30, mainly to distinguish large targets such as aircraft carriers, d...
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, Popular Technical Reports.
© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com