Integrating Multi-Source Transfer Learning, Active Learning and Metric Learning paradigms for Time Series Prediction

Qitao Gu, Qun Dai, HuiHui Yu, Rui Ye

Abstract: Highlights•This work proposes MS-AMTL algorithm, integrating MSTL, AL, and ML paradigms.•MS-AMTL transfers knowledge in a long-time span effectively.•MS-AMTL fully considers subtle changes in data distributions.•A density weighted uncertain sample selection scheme is designed in MS-AMTL.•MS-AMTL employs ML to take advantage of internal geometric information.