Batch active learning for time-series classification with multi-mode exploration

Sangho Lee, Chihyeon Choi, Hyungrok Do, Youngdoo Son

Abstract: Highlights•An efficient active learning method for time-series classification is proposed.•We consider multi-modality and in-batch diversity through distance distribution.•Two novel metrics enable effective multi-mode exploration and exploitation.•We theoretically show how the proposed method encourages in-batch diversity.•Our method performs superiorly regardless of the amount of class information.