Convolutional neural network-oriented collaborative sparse quantization method

By employing a structured collaborative sparse quantization method, which combines sparsity and quantization for collaborative optimization during training, the problem of error accumulation and accuracy degradation in convolutional neural networks is solved. This achieves efficient model compression and accuracy preservation, making it suitable for edge devices.

CN121189392APending Publication Date: 2025-12-23BEIJING INST OF TECH
0 Cites 0 Cited by

Patent Information

Application Number
CN202511339477.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-09-18
Publication Date
2025-12-23

AI Technical Summary

Technical Problem

Existing sparse pruning and quantization methods suffer from error accumulation and accuracy degradation in convolutional neural networks, especially when deployed on edge devices, making it difficult to balance model compression and performance maintenance.

Method used

A structured collaborative sparse quantization method is proposed. Through error theory analysis, structured sparsity construction strategy and logarithmic domain block quantization method, sparsity and quantization are optimized collaboratively during training. The influence of outliers is mitigated by utilizing the correlation of convolution kernel parameters and logarithmic domain mapping technique.

Benefits of technology

It achieves simultaneous sparsity and quantization during training, reducing model lightweighting compression time and resource overhead, improving model compression ratio, and maintaining high accuracy in detection tasks, achieving a compression effect of more than ten times.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses a collaborative sparse quantization method for a convolutional neural network, and the method comprises the three steps: carrying out the theoretical analysis of a collaborative sparse quantization error, and carrying out the main analysis of the collaborative sparse quantization error; the step 2 is a collaborative sparse method based on channel correlation, and the part is dominant; the step 3 is a block logarithm domain mapping average quantization method, and the step 3 is mainly used.
Need to check novelty before this filing date? Find Prior Art