Scholarlink

<p>In the fields of machine learning, deep learning, and statistical learning, many critical scientific problems can be described by a special class of optimization models based on regularization terms. However, how to choose an appropriate regularization term is a crucial problem. This paper employs phase diagram analysis to conduct a comprehensive experimental evaluation of eight prevalent sparse regularization models. The experiments primarily focus on recovering sparse signals generated from four distributions with different parameters in probability theory and statistics. The evaluation criteria such as recovery efficiency, recovery time, recovery stability and comprehensive metrics are considered comprehensively metrics are considered. To verify the accuracy of the phase diagram experiment, we conduct a logistic regression experiment, selecting four groups of actual data with different sizes, and analyze the test error and the CPU time. Finally, we analyze the performance characteristics, advantages and disadvantages of different regularization models, which provide theoretical guidance for users to choose.</p>

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

Performance Evaluation and Analysis of Sparse Regularization Based on Phase Diagram

  • 期刊论文

  • 发表日期：

    2025年8月18日

查看全文
  • Yi-Jie Li,
  • Qian Li,
  • Xue-Rui Gao,
  • Guo-Qiang Wang

摘要

In the fields of machine learning, deep learning, and statistical learning, many critical scientific problems can be described by a special class of optimization models based on regularization terms. However, how to choose an appropriate regularization term is a crucial problem. This paper employs phase diagram analysis to conduct a comprehensive experimental evaluation of eight prevalent sparse regularization models. The experiments primarily focus on recovering sparse signals generated from four distributions with different parameters in probability theory and statistics. The evaluation criteria such as recovery efficiency, recovery time, recovery stability and comprehensive metrics are considered comprehensively metrics are considered. To verify the accuracy of the phase diagram experiment, we conduct a logistic regression experiment, selecting four groups of actual data with different sizes, and analyze the test error and the CPU time. Finally, we analyze the performance characteristics, advantages and disadvantages of different regularization models, which provide theoretical guidance for users to choose.