Electrostatic force regularization for neural structured pruning

doi:10.1016/j.ins.2025.123045

Electrostatic force regularization for neural structured pruning

2025-12-01

0

PRE

AI

A

Abdesselam Ferdi *

A

Abdelmalik Taleb-Ahmed

DOI：10.1016/j.ins.2025.123045

Original

Original request for help

Save

Abstract

En 中文

The demand for deploying deep convolutional neural networks (DCNNs) on resource-constrained devices for real-time applications remains substantial. However, existing state-of-the-art structured pruning methods often involve intricate implementations, require modifications to the original network architectures, and necessitate an extensive fine-tuning phase. To address these challenges, we introduce a novel approach that integrates the concept of electrostatic force from physics into the training process of DCNNs. The magnitude of this force is directly proportional to the product of the charges of the convolution filter and the source filter, and inversely proportional to the square of the distance between them. We applied the electrostatic-like force to the convolution filters, attracting those with opposite charges toward non-zero weights and repelling similar ones toward zero. Consequently, repelled filters are pruned as their weights approach zero, while attracted filters retain significant parameters that preserve essential information. Unlike conventional methods, our approach is straightforward to implement, does not require any architectural modifications, and simultaneously optimizes weights and ranks filter importance, all without the need for extensive fine-tuning. We validated the efficacy of our method on modern DCNN architectures using the MNIST, CIFAR, and ImageNet datasets, achieving competitive performance compared to existing structured pruning approaches.

Keywords:

DCNNs

Electrostatic force

Structured pruning

Regularization

Electrostatic force regularization for neural structured pruning

Abstract

Journal

Organization

Cited Papers