Multioutput Gaussian Process Modulated Poisson Processes for Event Prediction

Prediction of events such as part replacement and failure plays a critical role in reliability engineering. Event stream data are commonly observed manufacturing teleservice systems. Designing predictive models for individual units based on event streams is challenging an underexplored problem. In this work, we propose nonparametric prognostic framework individualized prediction the inhomogeneous Poisson processes with multivariate Gaussian convolution process (MGCP) prior intensity functions. The MGCP functions maps from similar historical to current unit under study which facilitates sharing information allows analysis flexible patterns. To facilitate inference, derive variational inference scheme learning estimation parameters resulting modulated model. Experimental results shown both synthetic well real-world fleet-based prediction.