Hybrid active learning for non-stationary streaming data with asynchronous labeling

摘要: Active learning enables supervised classifiers to learn using fewer labeled samples, by actively selecting samples for human labeling. Most Learning approaches can be categorized as pool-based or stream-based. Pool-based strategies select instances from the available pool of unlabeled data, evaluating each instance, whereas stream-based examine every instance in incoming stream data and decide sequentially whether they want that not. Stream-based enable ability adapt classifier model more quickly changes, while often exhibit better rates. In this paper, we propose a framework method Hybrid integrates harvest benefits both, streaming classification scenario where concept drift may prevalent, labeling is asynchronous. addition, (i) prioritized aggregation selection combine selected strategies, (ii) batch period adaptation dynamically change triggering pattern strategy based upon detection drift.