Scholarlink

Access to correct labels is crucial from the point of view of classifier learning. Unfortunately, the labeling process involves a high cost in many cases, so methods based only on partially labeled data are becoming increasingly popular. This paper presents selected methods of data labeling using active learning. Experimental studies conducted on four benchmark databases confirm the usefulness of the methods chosen and their superiority over object labeling using a random approach.

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

Analysis of Selected Deep Active Learning Algorithms

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Mariusz Topolski,
  • Katarzyna Topolska,
  • Jan Wasilewski,
  • Michał Woźniak

摘要

Access to correct labels is crucial from the point of view of classifier learning. Unfortunately, the labeling process involves a high cost in many cases, so methods based only on partially labeled data are becoming increasingly popular. This paper presents selected methods of data labeling using active learning. Experimental studies conducted on four benchmark databases confirm the usefulness of the methods chosen and their superiority over object labeling using a random approach.