Deep Echo State Networks with Uncertainty Quantification for Spatio-Temporal Forecasting. McDermott, P. L. & Wikle, C. K. arXiv:1806.10728 [cs, stat], September, 2018. arXiv: 1806.10728![Deep Echo State Networks with Uncertainty Quantification for Spatio-Temporal Forecasting [link]](https://bibbase.org/img/filetypes/link.svg "link")
Paper abstract bibtex Long-lead forecasting for spatio-temporal systems can often entail complex nonlinear dynamics that are difficult to specify it a priori. Current statistical methodologies for modeling these processes are often highly parameterized and thus, challenging to implement from a computational perspective. One potential parsimonious solution to this problem is a method from the dynamical systems and engineering literature referred to as an echo state network (ESN). ESN models use so-called \\textbackslashit reservoir computing\ to efficiently compute recurrent neural network (RNN) forecasts. Moreover, so-called "deep" models have recently been shown to be successful at predicting high-dimensional complex nonlinear processes, particularly those with multiple spatial and temporal scales of variability (such as we often find in spatio-temporal environmental data). Here we introduce a deep ensemble ESN (D-EESN) model. We present two versions of this model for spatio-temporal processes that both produce forecasts and associated measures of uncertainty. The first approach utilizes a bootstrap ensemble framework and the second is developed within a hierarchical Bayesian framework (BD-EESN). This more general hierarchical Bayesian framework naturally accommodates non-Gaussian data types and multiple levels of uncertainties. The methodology is first applied to a data set simulated from a novel non-Gaussian multiscale Lorenz-96 dynamical system simulation model and then to a long-lead United States (U.S.) soil moisture forecasting application.
@article{mcdermott_deep_2018,
title = {Deep {Echo} {State} {Networks} with {Uncertainty} {Quantification} for {Spatio}-{Temporal} {Forecasting}},
url = {http://arxiv.org/abs/1806.10728},
abstract = {Long-lead forecasting for spatio-temporal systems can often entail complex nonlinear dynamics that are difficult to specify it a priori. Current statistical methodologies for modeling these processes are often highly parameterized and thus, challenging to implement from a computational perspective. One potential parsimonious solution to this problem is a method from the dynamical systems and engineering literature referred to as an echo state network (ESN). ESN models use so-called \{{\textbackslash}it reservoir computing\} to efficiently compute recurrent neural network (RNN) forecasts. Moreover, so-called "deep" models have recently been shown to be successful at predicting high-dimensional complex nonlinear processes, particularly those with multiple spatial and temporal scales of variability (such as we often find in spatio-temporal environmental data). Here we introduce a deep ensemble ESN (D-EESN) model. We present two versions of this model for spatio-temporal processes that both produce forecasts and associated measures of uncertainty. The first approach utilizes a bootstrap ensemble framework and the second is developed within a hierarchical Bayesian framework (BD-EESN). This more general hierarchical Bayesian framework naturally accommodates non-Gaussian data types and multiple levels of uncertainties. The methodology is first applied to a data set simulated from a novel non-Gaussian multiscale Lorenz-96 dynamical system simulation model and then to a long-lead United States (U.S.) soil moisture forecasting application.},
urldate = {2019-12-12},
journal = {arXiv:1806.10728 [cs, stat]},
author = {McDermott, Patrick L. and Wikle, Christopher K.},
month = sep,
year = {2018},
note = {arXiv: 1806.10728},
keywords = {Computer Science - Machine Learning, Forecasting, Statistics - Machine Learning, long-lead forecasting}
}
Downloads: 0
{"_id":"2NAsSsef2t7LYeXf3","bibbaseid":"mcdermott-wikle-deepechostatenetworkswithuncertaintyquantificationforspatiotemporalforecasting-2018","authorIDs":[],"author_short":["McDermott, P. L.","Wikle, C. K."],"bibdata":{"bibtype":"article","type":"article","title":"Deep Echo State Networks with Uncertainty Quantification for Spatio-Temporal Forecasting","url":"http://arxiv.org/abs/1806.10728","abstract":"Long-lead forecasting for spatio-temporal systems can often entail complex nonlinear dynamics that are difficult to specify it a priori. Current statistical methodologies for modeling these processes are often highly parameterized and thus, challenging to implement from a computational perspective. One potential parsimonious solution to this problem is a method from the dynamical systems and engineering literature referred to as an echo state network (ESN). ESN models use so-called \\\\textbackslashit reservoir computing\\ to efficiently compute recurrent neural network (RNN) forecasts. Moreover, so-called \"deep\" models have recently been shown to be successful at predicting high-dimensional complex nonlinear processes, particularly those with multiple spatial and temporal scales of variability (such as we often find in spatio-temporal environmental data). Here we introduce a deep ensemble ESN (D-EESN) model. We present two versions of this model for spatio-temporal processes that both produce forecasts and associated measures of uncertainty. The first approach utilizes a bootstrap ensemble framework and the second is developed within a hierarchical Bayesian framework (BD-EESN). This more general hierarchical Bayesian framework naturally accommodates non-Gaussian data types and multiple levels of uncertainties. The methodology is first applied to a data set simulated from a novel non-Gaussian multiscale Lorenz-96 dynamical system simulation model and then to a long-lead United States (U.S.) soil moisture forecasting application.","urldate":"2019-12-12","journal":"arXiv:1806.10728 [cs, stat]","author":[{"propositions":[],"lastnames":["McDermott"],"firstnames":["Patrick","L."],"suffixes":[]},{"propositions":[],"lastnames":["Wikle"],"firstnames":["Christopher","K."],"suffixes":[]}],"month":"September","year":"2018","note":"arXiv: 1806.10728","keywords":"Computer Science - Machine Learning, Forecasting, Statistics - Machine Learning, long-lead forecasting","bibtex":"@article{mcdermott_deep_2018,\n\ttitle = {Deep {Echo} {State} {Networks} with {Uncertainty} {Quantification} for {Spatio}-{Temporal} {Forecasting}},\n\turl = {http://arxiv.org/abs/1806.10728},\n\tabstract = {Long-lead forecasting for spatio-temporal systems can often entail complex nonlinear dynamics that are difficult to specify it a priori. Current statistical methodologies for modeling these processes are often highly parameterized and thus, challenging to implement from a computational perspective. One potential parsimonious solution to this problem is a method from the dynamical systems and engineering literature referred to as an echo state network (ESN). ESN models use so-called \\{{\\textbackslash}it reservoir computing\\} to efficiently compute recurrent neural network (RNN) forecasts. Moreover, so-called \"deep\" models have recently been shown to be successful at predicting high-dimensional complex nonlinear processes, particularly those with multiple spatial and temporal scales of variability (such as we often find in spatio-temporal environmental data). Here we introduce a deep ensemble ESN (D-EESN) model. We present two versions of this model for spatio-temporal processes that both produce forecasts and associated measures of uncertainty. The first approach utilizes a bootstrap ensemble framework and the second is developed within a hierarchical Bayesian framework (BD-EESN). This more general hierarchical Bayesian framework naturally accommodates non-Gaussian data types and multiple levels of uncertainties. The methodology is first applied to a data set simulated from a novel non-Gaussian multiscale Lorenz-96 dynamical system simulation model and then to a long-lead United States (U.S.) soil moisture forecasting application.},\n\turldate = {2019-12-12},\n\tjournal = {arXiv:1806.10728 [cs, stat]},\n\tauthor = {McDermott, Patrick L. and Wikle, Christopher K.},\n\tmonth = sep,\n\tyear = {2018},\n\tnote = {arXiv: 1806.10728},\n\tkeywords = {Computer Science - Machine Learning, Forecasting, Statistics - Machine Learning, long-lead forecasting}\n}\n\n","author_short":["McDermott, P. L.","Wikle, C. K."],"key":"mcdermott_deep_2018","id":"mcdermott_deep_2018","bibbaseid":"mcdermott-wikle-deepechostatenetworkswithuncertaintyquantificationforspatiotemporalforecasting-2018","role":"author","urls":{"Paper":"http://arxiv.org/abs/1806.10728"},"keyword":["Computer Science - Machine Learning","Forecasting","Statistics - Machine Learning","long-lead forecasting"],"downloads":0},"bibtype":"article","biburl":"https://bibbase.org/zotero/carlosgogo","creationDate":"2019-12-13T10:16:40.597Z","downloads":0,"keywords":["computer science - machine learning","forecasting","statistics - machine learning","long-lead forecasting"],"search_terms":["deep","echo","state","networks","uncertainty","quantification","spatio","temporal","forecasting","mcdermott","wikle"],"title":"Deep Echo State Networks with Uncertainty Quantification for Spatio-Temporal Forecasting","year":2018,"dataSources":["8KsDsKJXTPZFnaHDn"]}