CNN Processing Method Using Survival Network Selection for Speed and Precision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Convolutional neural networks (CNNs) face challenges in achieving high-speed processing for recognition and authentication operations in embedded systems with limited resources, such as smartphones, without compromising performance.
Innovation Solution
The method involves selecting a 'survival network' through a high-speed convolution operation, followed by a precision convolution operation using the survival network, which includes performing high-speed convolution operations by simplifying or omitting certain operations, and then supplementing accuracy using the precision convolution operation to enhance both speed and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If high-speed convolution operation is performed by simplifying or omitting certain operations, then processing speed is improved, but calculation precision deteriorates
Solution Approach 1:
The patent divides the convolution operation into two distinct stages: a high-speed convolution operation that performs simplified calculations to generate approximate output, and a precision convolution operation that performs accurate calculations only on selected elements. This segmentation allows the system to achieve both high processing speed and high calculation precision by applying different processing strategies to different parts of the data.
2Measurement precision
If precision convolution operation is performed on all elements, then calculation precision is improved, but processing time increases
Solution Approach 1:
The patent applies partial action by performing the computationally intensive precision convolution operation only on a subset of output elements that are likely to be used, rather than on all elements. The system identifies and selects only the necessary elements based on the approximate output from the high-speed convolution, thereby reducing processing time while maintaining calculation precision where it matters most.
3Productivity
If high-speed convolution operation is performed, then productivity is improved, but manufacturing precision deteriorates
Solution Approach 1:
The patent performs a preliminary high-speed convolution operation first to generate approximate output values. Based on these approximate results, the system then selectively performs precision convolution operations only on elements that require accurate calculations. This preliminary action approach allows the system to quickly identify which elements need precise processing, thereby improving overall productivity while maintaining output accuracy for critical elements.
Data Source
AI summary
A convolutional neural network (CNN) processing method includes selecting a survival network in a precision convolutional network based on a result of performing a high speed convolution operation between an input and a kernel using a high speed convolutional network, and performing a precision convolution operation between the input and the kernel using the survival network.