Fast Kronecker Inference in Gaussian Processes with non-Gaussian Likelihoods

Seth Flaxman, Andrew Wilson, Daniel Neill, Hannes Nickisch, Alex Smola
; Proceedings of the 32nd International Conference on Machine Learning, PMLR 37:607-616, 2015.

Abstract

Gaussian processes (GPs) are a flexible class of methods with state of the art performance on spatial statistics applications. However, GPs require O(n^3) computations and O(n^2) storage, and popular GP kernels are typically limited to smoothing and interpolation. To address these difficulties, Kronecker methods have been used to exploit structure in the GP covariance matrix for scalability, while allowing for expressive kernel learning (Wilson et al., 2014). However, fast Kronecker methods have been confined to Gaussian likelihoods. We propose new scalable Kronecker methods for Gaussian processes with non-Gaussian likelihoods, using a Laplace approximation which involves linear conjugate gradients for inference, and a lower bound on the GP marginal likelihood for kernel learning. Our approach has near linear scaling, requiring O(D n^(D+1)/D) operations and O(D n^2/D) storage, for n training data-points on a dense D > 1 dimensional grid. Moreover, we introduce a log Gaussian Cox process, with highly expressive kernels, for modelling spatiotemporal count processes, and apply it to a point pattern (n = 233,088) of a decade of crime events in Chicago. Using our model, we discover spatially varying multiscale seasonal trends and produce highly accurate long-range local area forecasts.

Cite this Paper


BibTeX
@InProceedings{pmlr-v37-flaxman15, title = {Fast Kronecker Inference in Gaussian Processes with non-Gaussian Likelihoods}, author = {Seth Flaxman and Andrew Wilson and Daniel Neill and Hannes Nickisch and Alex Smola}, booktitle = {Proceedings of the 32nd International Conference on Machine Learning}, pages = {607--616}, year = {2015}, editor = {Francis Bach and David Blei}, volume = {37}, series = {Proceedings of Machine Learning Research}, address = {Lille, France}, month = {07--09 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v37/flaxman15.pdf}, url = {http://proceedings.mlr.press/v37/flaxman15.html}, abstract = {Gaussian processes (GPs) are a flexible class of methods with state of the art performance on spatial statistics applications. However, GPs require O(n^3) computations and O(n^2) storage, and popular GP kernels are typically limited to smoothing and interpolation. To address these difficulties, Kronecker methods have been used to exploit structure in the GP covariance matrix for scalability, while allowing for expressive kernel learning (Wilson et al., 2014). However, fast Kronecker methods have been confined to Gaussian likelihoods. We propose new scalable Kronecker methods for Gaussian processes with non-Gaussian likelihoods, using a Laplace approximation which involves linear conjugate gradients for inference, and a lower bound on the GP marginal likelihood for kernel learning. Our approach has near linear scaling, requiring O(D n^(D+1)/D) operations and O(D n^2/D) storage, for n training data-points on a dense D > 1 dimensional grid. Moreover, we introduce a log Gaussian Cox process, with highly expressive kernels, for modelling spatiotemporal count processes, and apply it to a point pattern (n = 233,088) of a decade of crime events in Chicago. Using our model, we discover spatially varying multiscale seasonal trends and produce highly accurate long-range local area forecasts.} }
Endnote
%0 Conference Paper %T Fast Kronecker Inference in Gaussian Processes with non-Gaussian Likelihoods %A Seth Flaxman %A Andrew Wilson %A Daniel Neill %A Hannes Nickisch %A Alex Smola %B Proceedings of the 32nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2015 %E Francis Bach %E David Blei %F pmlr-v37-flaxman15 %I PMLR %J Proceedings of Machine Learning Research %P 607--616 %U http://proceedings.mlr.press %V 37 %W PMLR %X Gaussian processes (GPs) are a flexible class of methods with state of the art performance on spatial statistics applications. However, GPs require O(n^3) computations and O(n^2) storage, and popular GP kernels are typically limited to smoothing and interpolation. To address these difficulties, Kronecker methods have been used to exploit structure in the GP covariance matrix for scalability, while allowing for expressive kernel learning (Wilson et al., 2014). However, fast Kronecker methods have been confined to Gaussian likelihoods. We propose new scalable Kronecker methods for Gaussian processes with non-Gaussian likelihoods, using a Laplace approximation which involves linear conjugate gradients for inference, and a lower bound on the GP marginal likelihood for kernel learning. Our approach has near linear scaling, requiring O(D n^(D+1)/D) operations and O(D n^2/D) storage, for n training data-points on a dense D > 1 dimensional grid. Moreover, we introduce a log Gaussian Cox process, with highly expressive kernels, for modelling spatiotemporal count processes, and apply it to a point pattern (n = 233,088) of a decade of crime events in Chicago. Using our model, we discover spatially varying multiscale seasonal trends and produce highly accurate long-range local area forecasts.
RIS
TY - CPAPER TI - Fast Kronecker Inference in Gaussian Processes with non-Gaussian Likelihoods AU - Seth Flaxman AU - Andrew Wilson AU - Daniel Neill AU - Hannes Nickisch AU - Alex Smola BT - Proceedings of the 32nd International Conference on Machine Learning PY - 2015/06/01 DA - 2015/06/01 ED - Francis Bach ED - David Blei ID - pmlr-v37-flaxman15 PB - PMLR SP - 607 DP - PMLR EP - 616 L1 - http://proceedings.mlr.press/v37/flaxman15.pdf UR - http://proceedings.mlr.press/v37/flaxman15.html AB - Gaussian processes (GPs) are a flexible class of methods with state of the art performance on spatial statistics applications. However, GPs require O(n^3) computations and O(n^2) storage, and popular GP kernels are typically limited to smoothing and interpolation. To address these difficulties, Kronecker methods have been used to exploit structure in the GP covariance matrix for scalability, while allowing for expressive kernel learning (Wilson et al., 2014). However, fast Kronecker methods have been confined to Gaussian likelihoods. We propose new scalable Kronecker methods for Gaussian processes with non-Gaussian likelihoods, using a Laplace approximation which involves linear conjugate gradients for inference, and a lower bound on the GP marginal likelihood for kernel learning. Our approach has near linear scaling, requiring O(D n^(D+1)/D) operations and O(D n^2/D) storage, for n training data-points on a dense D > 1 dimensional grid. Moreover, we introduce a log Gaussian Cox process, with highly expressive kernels, for modelling spatiotemporal count processes, and apply it to a point pattern (n = 233,088) of a decade of crime events in Chicago. Using our model, we discover spatially varying multiscale seasonal trends and produce highly accurate long-range local area forecasts. ER -
APA
Flaxman, S., Wilson, A., Neill, D., Nickisch, H. & Smola, A.. (2015). Fast Kronecker Inference in Gaussian Processes with non-Gaussian Likelihoods. Proceedings of the 32nd International Conference on Machine Learning, in PMLR 37:607-616

Related Material