Variational Gaussian process for optimal sensor placement

Sensor placement is an optimisation problem that has recently gained great relevance. In order to achieve accurate online updates of a predictive model, sensors are used provide observations. When sensor location optimally selected, the model can greatly reduce its internal errors. A greedy-selection algorithm for locating these optimal spatial locations from numerical embedded space. novel architecture solving this big data proposed, relying on variational Gaussian process. The generalisation further improved via preconditioning inputs: Masked Autoregressive Flows implemented learn nonlinear, invertible transformations conditionally modelled features. Finally, global strategy extending Mutual Information-based and fine-tuning selected proposed. methodology parallelised speed up computational time, making tools very fast despite high complexity associated with both modelling tasks. applied real three-dimensional test case considering room within Clarence Centre building located in Elephant Castle, London, UK.