Scalable Gaussian Process Classification with Additive Noise for Various Likelihoods

--Gaussian process classification (GPC) provides a flexible and powerful statistical framework describing joint distributions over function space. Conventional GPCs however suffer from (i) poor scalability for big data due to the full kernel matrix, and (ii) intractable inference due to the non-Gaussian likelihoods. Hence, various scalable GPCs have been proposed through (i) the sparse approximation built upon a small inducing set to reduce the time complexity; and (ii) the approximate inference to derive analytical evidence lower bound (ELBO). However, these scalable GPCs equipped with analytical ELBO are limited to specific likelihoods or additional assumptions. In this work, we present a unifying framework which accommodates scalable GPCs using various likelihoods. Analogous to GP regression (GPR), we introduce additive noises to augment the probability space for (i) the GPCs with step, (multinomial) probit and logit likelihoods via the internal variables; and particularly, (ii) the GPC using softmax likelihood via the noise variables themselves. This leads to unified scalable GPCs with analytical ELBO by using variational inference. S a nonparametric Bayesian model which is explainable and provides confidence in predictions, Gaussian process (GP) has been widely investigated and used in various scenarios, e.g., regression and classification [1], active learning [2], unsupervised learning [3], and multi-task learning [4], [5]. The central task in GP is to infer the latent function f, which follows a Gaussian process GP (0,k (.)) where the kernel k ( .) It is more challenging than the GP regression (GPR).