Split ‘n’ merge net: A dynamic masking network for multi-task attention

作者：

Highlights：

• A novel outlook on multi-task learning as a feature separation problem.

• A simplified, interpretable, and dynamic framework for multi-task learning.

• Multi-head attention based dynamic feature masking to select task specific and shared features from an embedding.

• Achieves state-of-the-art results in multiple public benchmarks.

摘要

•A novel outlook on multi-task learning as a feature separation problem.•A simplified, interpretable, and dynamic framework for multi-task learning.•Multi-head attention based dynamic feature masking to select task specific and shared features from an embedding.•Achieves state-of-the-art results in multiple public benchmarks.

论文关键词：Multi-task learning,Attention,Cuffless blood pressure measurement,Biomedical signal processing,Deep learning,Emotion recognition

论文评审过程：Received 12 September 2021, Revised 21 December 2021, Accepted 22 January 2022, Available online 25 January 2022, Version of Record 30 January 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108551