Scholarlink

This study aims to reduce the computational complexity of deep learning models through structured pruning. It introduces a channel pruning approach based on sparse learning, designed to simplify networks while maintaining high performance. By enforcing channel-level sparsity in an efficient yet straightforward manner, the method identifies and removes insignificant channels during training, transforming wide and large networks into compact, efficient models with minimal accuracy loss. Experimental results on modern architectures across multiple benchmark datasets confirm the effectiveness of this approach, demonstrating significant model compression while preserving competitive accuracy compared to state-of-the-art techniques.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Efficient Channel Pruning via Sparse Learning for Compact Deep Neural Networks

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Khalid Elghazi,
  • Hassan Ramchoun,
  • Tawfik Masrour

摘要

This study aims to reduce the computational complexity of deep learning models through structured pruning. It introduces a channel pruning approach based on sparse learning, designed to simplify networks while maintaining high performance. By enforcing channel-level sparsity in an efficient yet straightforward manner, the method identifies and removes insignificant channels during training, transforming wide and large networks into compact, efficient models with minimal accuracy loss. Experimental results on modern architectures across multiple benchmark datasets confirm the effectiveness of this approach, demonstrating significant model compression while preserving competitive accuracy compared to state-of-the-art techniques.