Variational Tobit Gaussian Process Regression

Abstract We propose a variational inference-based framework for training Gaussian process regression model subject to censored observational data. Data censoring is typical problem encountered during the data gathering procedure and requires specialized techniques perform inference since resulting probabilistic models are typically analytically intractable. In this article we exploit sparse inducing variable local methods compute an tractable lower bound on true log marginal likelihood of which can be used Bayesian inference. demonstrate proposed synthetically-produced, noise-corrupted data, as well real-world set, artificial censoring. The predictions comparable existing account censoring, but provides significant reduction in computational cost.