Nonmyopic (cid:15) -Bayes-Optimal Active Learning of Gaussian Processes
Nonmyopic (cid:15) -Bayes-Optimal Active Learning of Gaussian Processes
复制标题
非近视 (cid:15) - 高斯过程的贝叶斯最优主动学习
DOI：
10.1609/aaai.v31i1.10772
复制
发表时间：
2014
期刊：
影响因子：
--
通讯作者：
†. MohanKankanhalli
中科院分区：
文献类型：
--
作者：
Trong Nghia;Hoang;†. KianHsiangLow;§. PatrickJaillet;†. MohanKankanhalli
高斯过程主动学习的一个基本问题是探索与利用的权衡。本文提出了一种新颖的非近视 (cid:15) -贝叶斯最优主动学习 ((cid:15) -BAL) 方法，该方法可以联合且自然地优化权衡。相比之下，现有的工作主要是开发近视/贪婪算法或分别进行探索和利用。为了实时执行主动学习，我们提出了一种基于 (cid:15) -BAL 的随时算法，具有性能保证，并使用合成和真实数据集进行实证证明，在预算有限的情况下，它的性能优于最先进的算法。
A fundamental issue in active learning of Gaussian processes is that of the exploration-exploitation trade-off. This paper presents a novel nonmyopic (cid:15) -Bayes-optimal active learning ( (cid:15) -BAL) approach that jointly and naturally optimizes the trade-off. In contrast, existing works have primarily developed myopic/greedy algorithms or performed exploration and exploitation separately. To perform active learning in real time, we then propose an anytime algo-rithm based on (cid:15) -BAL with performance guarantee and empirically demonstrate using synthetic and real-world datasets that, with limited budget, it outperforms the state-of-the-art algorithms.