Relation-preserving masked modeling for semi-supervised time-series classification

Sangho Lee, Chihyeon Choi, Youngdoo Son

Abstract: Highlights•We propose a novel masked modeling for semi-supervised time-series classification.•A dual-temporal encoder is designed to reflect diverse temporal resolutions.•We introduce a novel loss function to mitigate information loss within the encoder.•We use random masking ratios to boost model performance without exploring optima.•Our method outperforms the baselines in semi-supervised time-series classification.