Spatial Dynamics Complementarity Learning with Graph Convolutional Network for Wearable Massive-sensor Computers.

Wearable massive-sensor computers show great promise for comprehensive bio-dynamics capturing towards smart health big data. Nevertheless, mining the critical and complementary patterns from a large amount of data, especially with a high dimensionality, is challenging. In this study, we propose a novel deep learning framework, with both deep graph convolutional learning and multi-view representation, to boost the spatial dynamics complementarity mining. More specifically, we have proposed a multi-stage deep learning architecture, with a graph convolutional network for spatial pattern propagation and fusion, and a convolutional network for further pattern abstraction. Further, we have proposed a multi-view data representation approach with both temporal and spectral feature formation. Evaluated on the wearable massive-sensor data mining application, the proposed framework has greatly boosted the complementarity learning capability and yielded a significantly improved detection performance. This study is expected to greatly advance spatial dynamics complementarity learning with graph convolutional learning.