Scholarlink

Besides adding sparsity-inducing regularization to the optimization objective, sparsity can also be achieved by directly altering the optimization algorithm. This chapter will first introduce the concept of proximal gradient descent and proximal operators, then cover optimization methods specifically designed for DNN models like continuous sparsification, gradient-based pruning criteria, and dynamic DNN models.

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

Optimization Techniques for Sparsity

  • 章节

  • 发表日期：

    2026年

查看全文
  • Yiran Chen,
  • Hai Li,
  • Huanrui Yang

摘要

Besides adding sparsity-inducing regularization to the optimization objective, sparsity can also be achieved by directly altering the optimization algorithm. This chapter will first introduce the concept of proximal gradient descent and proximal operators, then cover optimization methods specifically designed for DNN models like continuous sparsification, gradient-based pruning criteria, and dynamic DNN models.