Smoothed per-tensor weight quantization: a robust solution for neural network deployment

Authors

  • Xin Chang Warsaw University of Technology, Poland

DOI:

https://doi.org/10.24425/ijet.2025.153629

Abstract

This paper introduces a novel method to improve quantization outcomes for per-tensor weight quantization, focusing on enhancing computational efficiency and compatibility with resource-constrained hardware. Addressing the inherent challenges of depth-wise convolutions, the proposed smooth quantization technique redistributes weight magnitude disparities to pre-activation data, thereby equalizing channel-wise weight magnitudes. This adjustment enables more effective application of uniform quantization schemes. Experimental evaluations on the ImageNet classification benchmark demonstrate substantial performance gains across modern architectures and training strategies. The proposed method achieves improved accuracy to per-tensor quantization without noticeable computational overhead, making it a practical solution for edge-device deployments.

Published

2025-07-11

How to Cite

Chang, Xin. “Smoothed Per-Tensor Weight Quantization: A Robust Solution for Neural Network Deployment”. International Journal of Electronics and Telecommunications, vol. 71, no. 3, July 2025, doi:10.24425/ijet.2025.153629.

Issue

Section

Artykuły