An enhanced category detection based on active learning

Identification of useful anomalies is an emerging task in active learning scenario. It plays the central roles in category detection in which one can using a sampling approach to label a data from rare category in an unlabeled date set by the help of the oracle who has a small querying budget. This paper presents an enhanced category detection that improves previous research work which leans to cost more querying budget. The new approach takes full advantage of the feedback of the oracle, and reduces the querying times. Experimental results on both synthetic and real data sets are effective and low-cost.