Exploring Vision Neural Network Pruning via Screening Methodology

arXiv stat.ML / 5/1/2026

💬 OpinionIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper addresses the practical cost of large deep neural networks by proposing a pruning framework that can cut both storage and computation by about an order of magnitude without significantly hurting accuracy.
It identifies non-essential parameters by statistically analyzing component significance across classification categories, using an F-statistic-based screening method.
The method combines F-statistic screening with a weighted evaluation scheme to estimate the contribution of connections and channels, supporting both unstructured pruning and structured pruning in one unified approach.
Experiments on real-world vision datasets show results on both fully connected (FNN) and convolutional (CNN) networks, with pruned models that remain competitive with state-of-the-art pruning techniques.
The overall goal is to produce compact, efficient models suitable for deployment on resource-constrained platforms like edge devices requiring real-time and energy-efficient inference.

Abstract

The remarkable performance of modern deep neural networks (DNNs) is largely driven by their massive scale, often comprising tens to hundreds of millions-or even billions-of parameters. However, such a scale incurs substantial storage and computational costs, hindering deployment on platforms such as edge devices that require energy-efficient and real-time processing. In this paper, we propose a network pruning framework that reduces both storage and computation requirements by an order of magnitude while preserving model accuracy. Our approach eliminates non-essential parameters through a statistical analysis of component significance across classification categories. Specifically, we employ a F-statistic-based screening technique combined with a weighted evaluation scheme to quantify the contributions of connections and channels, enabling both unstructured and structured pruning within a unified framework. Extensive experiments on real-world vision datasets, covering both fully connected neural networks (FNNs) and convolutional neural networks (CNNs), demonstrate that the proposed framework produces compact and efficient models that are highly competitive with the state of art apporoaches.