AITOP

$TCP_α$

共 1 条相关 AI 资讯
8月21日
10:46
10:46官方账号arXiv cs.LG@Parampreet Singh, Anushka Singh, Sumit Kumar, Vipul Arora
Proposes $TCP_α$, a novel confidence target for music information retrieval, improving accuracy by introducing a margin-controlled penalty for misclassified samples. Achieves macro-F1 score of 0.98 by rejecting 8% of least-confident predictions and restoring performance under domain shift with 5% labeled samples from a new corpus.
论文confidence estimationmusic information retrieval$TCP_α$
推荐理由：Read this if you're interested in improving music information retrieval accuracy with novel confidence targets like $TCP_α$, which outperforms existing methods and can handle domain shifts effectively.
原文