Distributed Regression Towards In-network Data Modeling for Wireless Sensor Networks

Data modelling is a potential way of leveraging data spatial-temporal correlations for energy-saving data acquisition in wireless sensor networks. This project focuses on developing a methodology of distributed data regression. The distributed data representation in the form of linear kernel regression is investigated in terms of kernel parameterizations. The in-network implementation of performing the optimality for distributed modelling is studied by combining the clustering based hierarchical routing tree with the distributed representation. The proposed methods promise more robust modelling yet less energy expense. The output of the project will be a complete solution for efficient data modelling with a detailed in-network protocol, which can promote wider applications such as distributed inference and other data query processing.