Neural Network Weight Classification for Lower Calculation Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing calculation amount in deep neural networks poses a challenge that needs to be addressed to improve efficiency and reduce computational resources.
Innovation Solution
A data processing method that involves classifying weights and feature data, performing calculations based on classification features, and adjusting network dimensions to optimize resource utilization, thereby reducing the calculation amount.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep neural network with many hidden layers is constructed to improve classification accuracy, then classification accuracy is improved, but calculation amount increases
Solution Approach 1:
The patent segments the neural network weights into multiple groups based on their absolute value magnitudes. Each group is processed separately with optimized calculation methods, reducing the overall calculation amount while maintaining classification accuracy through systematic weight-wise processing
Solution Approach 2:
The patent changes the calculation parameters by using different calculation methods for different weight groups. Specifically, it applies sign-bit based calculation for some groups and magnitude-based calculation for others, optimizing the balance between accuracy and computational efficiency
2Measurement precision
If deep neural network with many hidden layers is constructed to improve classification accuracy, then classification accuracy is improved, but energy consumption increases
Solution Approach 1:
The patent segments weights into multiple groups and processes each group with optimized calculation methods. This segmentation reduces redundant calculations and energy consumption while maintaining the necessary classification accuracy through systematic processing of weight groups
Solution Approach 2:
The patent applies different calculation parameters and methods for different weight groups, using sign-bit based and magnitude-based calculations selectively. This parameter optimization reduces energy consumption by avoiding unnecessary precise calculations where they are not needed
Data Source
AI summary
Provided are a data processing method and device, and a computer-readable storage medium. The data processing method includes: acquiring at least one type of weights and feature data corresponding to each type of the weights; acquiring, according to the at least one type of weights, a classification feature corresponding to each type of the weights; performing calculation according to the at least one type of weights and the feature data corresponding to each type of the weights to obtain a first processing result corresponding to each type of the weights; and performing calculation according to the classification feature and the first processing result corresponding to each type of the weights corresponding to each type of the weights to obtain a second processing result.