Assessing Approximate Inference for Binary Gaussian Process Classification

By Malte Kuss, Carl E. Rasmussen. Journal of machine learning research, 2005

346 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/6daefda840b79015b134aee52d6fb4903955a5d8

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

A Scalable Hierarchical Gaussian Process Classifier (2019, methodology): “GAUSSIAN process (GP) models are powerful nonparametric tools for Bayesian regression [1]–[4] and classification [5]–[9].”
Bayesian Gaussian Process Classification with the EM-EP Algorithm (2006, unknown): “Ç”
Gaussian Process Regression Networks (2011, background): “They are expressive, interpretable, avoid over-ﬁtting, and have impressive predictive performance in many thorough empirical comparisons (Rasmussen, 1996; Kuss and Rasmussen, 2005; Ras-mussen and Williams, 2006).”
A Unifying Framework for Sparse Gaussian Process Approximation using Power Expectation Propagation (2016, unknown): “In addition, learning pseudo-point input locations via optimisation of the model likelihood can perform poorly e.g. for DTC it is prone to overﬁtting even for M N [8].”
Expectation Propagation in Gaussian Process Dynamical Systems (2012, methodology): “Expectation Propagation [10, 11] is a widely-used deterministic algorithm for approximate Bayesian inference that has been shown to be highly accurate in many problems, including sparse regression models [17], GP classification [9], and inference in dynamical systems [13, 6, 18].”
Piecewise Bounds for Estimating Bernoulli-Logistic Latent Gaussian Models (2011, methodology): “Some authors instead advocate the use of a Jensen's inequality-based approximation to the marginal likelihood (Kuss & Rasmussen, 2005), which is identical to the one we apply.”
Elliptical slice sampling (2009, methodology): “We ran tests on the USPS classification problem as set up by Kuss and Rasmussen (2005). We used log σf =3.”
Gaussian Processes for Big Data through Stochastic Variational Inference (2012, methodology): “The usual procedure for fitting GPs with non-Gaussian likelihoods is to approximate the likelihood using either a local variational lower bound [4], or by expectation propagation [7].”
Robust Multi-Class Gaussian Process Classification (2011, methodology): “EP has shown good overall performance when compared to other methods in the task of classification with binary Gaussian processes [15, 16].”
Slice sampling covariance hyperparameters of latent Gaussian models (2010, methodology): “For classiﬁ-cation problems we used moment matching, because Laplace approximations do not work well [15]. surr-taylor: using surrogate data with noise variance set via Taylor expansion of the log-likelihood (13).”
Variational learning for latent Gaussian model of discrete data (2012, methodology): “3 Expectation Propagation Another approximation method, called expectation-propagation (EP) [Minka, 2001], has been applied extensively to LGMs [Kuss and Rasmussen, 2005; Nickisch and Rasmussen, 2008; Seeger and Nickisch, 2011; Zoeter et al., 2005].”
How to Explain Individual Classification Decisions (2009, methodology): “Because the posterior cannot be calculated analytically for GP classiﬁcation models, we used an approximation by expectation propagation (EP) (Kuss and Ramussen, 2005).”
Stochastic Expectation Propagation (2015, background): “At first sight, it therefore appears well suited to large-data problems: the locality of computation make the algorithm simple to parallelise and distribute, and good practical performance on a range of small data applications suggest that it will be accurate [9, 10, 11].”
Approximate Gaussian Integration using Expectation Propagation (2011, background): “Another important existing connection point is Gaussian Process classification (Rasmussen and Williams, 2006; Kuss and Rasmussen, 2005).”