Most likely heteroscedastic Gaussian process regression

By Kristian Kersting, Christian Plagemann, Patrick Pfaff, Wolfram Burgard. 2007

413 citations recorded by the source corpus.

Subfields: Analytical Chemistry, Artificial Intelligence, Control and Systems Engineering

How researchers see this paper

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

In Analytical Chemistry

In Artificial Intelligence

In Control and Systems Engineering

Citations in the corpus

Taking the Human Out of the Loop: A Review of Bayesian Optimization (2016, neutral): “, [86], [95], and”
Reinforcement Learning for Bandit Neural Machine Translation with Simulated Human Feedback (2017, background): “…question is whether we could extend our framework to model individual annotator preferences (Passonneau and Carpenter, 2014) or learn personalized models (Mirkin et al., 2015; Rabi-novich et al., 2017), and handle heteroscedastic noise (Park, 1966; Kersting et al., 2007; Antos et al., 2010).”
Assessment and Validation of Machine Learning Methods for Predicting Molecular Atomization Energies. (2013, background): “71,72 Regions of low data density Even if the data is taken from a single source, an explorative analysis of the data distribution is recommended.”
Gaussian Process Regression Networks (2011, background): “There has similarly been great interest in extending Gaussian process (GP) regression to account for input dependent noise variances (Goldberg et al., 1998; Kersting et al., 2007; Adams & Stegle, 2008; Turner, 2010; Wilson & Ghahramani, 2010b;a; Lázaro-Gredilla & Titsias, 2011).”
Recent Advances in Bayesian Optimization (2022, neutral): “[130] developed a maximum-a-posteriori approach.”
GP-BayesFilters: Bayesian filtering using Gaussian process prediction and observation models (2008, methodology): “Finally, heteroscedastic processes, where noise is state dependent, can be modeled using techniques found in (Kersting et al. 2007) and (Goldberg et al. 1998).”
Variational Bayesian Optimization for Runtime Risk-Sensitive Control (2012, neutral): “In the Bayesian setting, a second GP prior, g(θ) ∼ GP(μ0, kg(θ,θ)), is placed over the unknown log variance function, g(θ) ≡ log r(θ) [7, 10, 15].”
Optimization as Estimation with Gaussian Processes in Bandit Settings (2015, neutral): “functions sampled from heteroscedastic Gaussian processes [11, 20].”
Variational Heteroscedastic Gaussian Process Regression (2011, methodology): “There is relevant previous work on approximate inference in HGP regression, particularly the Markov chain Monte Carlo (MCMC) fully Bayesian method considered in (Goldberg et al., 1998) and the maximum a posteriori (MAP) approach used in (Kersting et al., 2007; Quadrianto et al., 2009).”
Learning Non-stationary System Dynamics Online Using Gaussian Processes (2010, methodology): “Throughout the experimental section, we use three different GP models: a standard GP model (StdGP) of [1], a heteroscedastic GP model (HetGP) of [3], and our GP model with individual, uncorrelated noise levels (InGP).”
A probabilistic data-driven model for planar pushing (2017, methodology): “[19] proposed a simplified learning algorithm based on maximum likelihood approach, which tends to underestimate noise levels.”
Learning GP-BayesFilters via Gaussian process latent variable models (2009, methodology): “For example, sparse GPs or heteroscedastic GPs (Kersting et al. 2007), ones where the noise predictions are state dependent, can be used at this time.”
Robot Learning by Demonstration with local Gaussian process regression (2010, methodology): “After a short review we extend the standard Gaussian process model using heteroscedasticity [10], [13].”
A Tutorial on Bayesian Optimization (2018, background): “Gaussian processes (or variants thereof such as Snoek et al. (2014) and Kersting et al. (2007)) are used in most work on Bayesian optimization, but it seems likely that classes of problems exist where the objective could be better modeled through other approaches.”
GPS-ABC: Gaussian Process Surrogate Approximate Bayesian Computation (2014, methodology): “Improvements to our GPs may be achieved using an inputdependent noise model [11, 13], where the noise is an additional independent Gaussian process.”