1、For such applications, active learning methods provide solutions that optimize the labeling process by selecting the most useful unlabeled instances to be passed to an oracle for labeling. Thereby, active learning aims to achieve high performance with as few labeled instances as possible.
2. Active learning is an emerging machine learning technique, which works with fewer labeled instances, thereby reducing labeling as well as learning cost.
3. The few works that have addressed this task so far mostly use either uncertainty sampling for active learning with support vector machines, thereby concentrating on instances close to the anticipated decision boundary
4. Following the recently proposed probabilistic active learning framework [17], the key idea is to compute the expectation over the true posterior by incorporating the number of labels in a neighborhood of the label candidate as a proxy for the posterior’s reliability
根据最近提出的概率主动学习框架[17],关键思想是通过将标签候选区域内的标签数量作为后验可靠性的代理,计算对真实后验的期望