Scholarlink

Active Learning aims to achieve a accuracy based heuristic within a specified labeling budget. We leverage self-supervised features as a feature differentiation paradigm and introduce a dynamic, density-based sampler to ensure diverse data point selection. Our framework combines these features with the sampler to identify challenging to classify samples while maintaining diversity. It outperforms established Active Learning baselines by 1.39% and demonstrates robustness across diverse datasets. By calibrating the sampling process, our method offers a straightforward yet effective solution to the cold start problem.

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

A Density-Based Active Learning Framework Leveraging Self-supervised Features

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Punit Kumar,
  • Shubhendu Sharma

摘要

Active Learning aims to achieve a accuracy based heuristic within a specified labeling budget. We leverage self-supervised features as a feature differentiation paradigm and introduce a dynamic, density-based sampler to ensure diverse data point selection. Our framework combines these features with the sampler to identify challenging to classify samples while maintaining diversity. It outperforms established Active Learning baselines by 1.39% and demonstrates robustness across diverse datasets. By calibrating the sampling process, our method offers a straightforward yet effective solution to the cold start problem.