Multivariate Time Series Forecasting with Dynamic Graph Neural ODEs. Jin, M., Zheng, Y., Li, Y., Chen, S., Yang, B., & Pan, S. https://github.com/GRAND-Lab/MTGODE, 2022. cite arxiv:2202.08408Comment: 14 pages, 6 figures, 5 tablesPaper doi abstract bibtex Multivariate time series forecasting has long received significant attention in real-world applications, such as energy consumption and traffic prediction. While recent methods demonstrate good forecasting abilities, they have three fundamental limitations. (i) Discrete neural architectures: Interlacing individually parameterized spatial and temporal blocks to encode rich underlying patterns leads to discontinuous latent state trajectories and higher forecasting numerical errors. (ii) High complexity: Discrete approaches complicate models with dedicated designs and redundant parameters, leading to higher computational and memory overheads. (iii) Reliance on graph priors: Relying on predefined static graph structures limits their effectiveness and practicability in real-world applications. In this paper, we address all the above limitations by proposing a continuous model to forecast $\textbf{M}$ultivariate $\textbf{T}$ime series with dynamic $\textbf{G}$raph neural $\textbf{O}$rdinary $\textbf{D}$ifferential $\textbf{E}$quations ($\texttt{MTGODE}$). Specifically, we first abstract multivariate time series into dynamic graphs with time-evolving node features and unknown graph structures. Then, we design and solve a neural ODE to complement missing graph topologies and unify both spatial and temporal message passing, allowing deeper graph propagation and fine-grained temporal information aggregation to characterize stable and precise latent spatial-temporal dynamics. Our experiments demonstrate the superiorities of $\texttt{MTGODE}$ from various perspectives on five time series benchmark datasets.
@misc{jin2022multivariate,
abstract = {Multivariate time series forecasting has long received significant attention
in real-world applications, such as energy consumption and traffic prediction.
While recent methods demonstrate good forecasting abilities, they have three
fundamental limitations. (i) Discrete neural architectures: Interlacing
individually parameterized spatial and temporal blocks to encode rich
underlying patterns leads to discontinuous latent state trajectories and higher
forecasting numerical errors. (ii) High complexity: Discrete approaches
complicate models with dedicated designs and redundant parameters, leading to
higher computational and memory overheads. (iii) Reliance on graph priors:
Relying on predefined static graph structures limits their effectiveness and
practicability in real-world applications. In this paper, we address all the
above limitations by proposing a continuous model to forecast
$\textbf{M}$ultivariate $\textbf{T}$ime series with dynamic $\textbf{G}$raph
neural $\textbf{O}$rdinary $\textbf{D}$ifferential $\textbf{E}$quations
($\texttt{MTGODE}$). Specifically, we first abstract multivariate time series
into dynamic graphs with time-evolving node features and unknown graph
structures. Then, we design and solve a neural ODE to complement missing graph
topologies and unify both spatial and temporal message passing, allowing deeper
graph propagation and fine-grained temporal information aggregation to
characterize stable and precise latent spatial-temporal dynamics. Our
experiments demonstrate the superiorities of $\texttt{MTGODE}$ from various
perspectives on five time series benchmark datasets.},
added-at = {2023-04-12T09:42:14.000+0200},
author = {Jin, Ming and Zheng, Yu and Li, Yuan-Fang and Chen, Siheng and Yang, Bin and Pan, Shirui},
biburl = {https://www.bibsonomy.org/bibtex/223ded3d3d60124236732b3deb0609001/manli},
description = {Multivariate Time Series Forecasting with Dynamic Graph Neural ODEs},
doi = {10.1109/TKDE.2022.3221989},
howpublished = {https://github.com/GRAND-Lab/MTGODE},
interhash = {62fd3e28a424edddc6265a90fe9b54a4},
intrahash = {23ded3d3d60124236732b3deb0609001},
keywords = {ma_sem2023},
note = {cite arxiv:2202.08408Comment: 14 pages, 6 figures, 5 tables},
timestamp = {2023-04-12T09:43:59.000+0200},
title = {Multivariate Time Series Forecasting with Dynamic Graph Neural ODEs},
url = {http://arxiv.org/abs/2202.08408},
year = 2022
}
Downloads: 0
{"_id":"56FnWWaSZbARFqnPp","bibbaseid":"jin-zheng-li-chen-yang-pan-multivariatetimeseriesforecastingwithdynamicgraphneuralodes-2022","author_short":["Jin, M.","Zheng, Y.","Li, Y.","Chen, S.","Yang, B.","Pan, S."],"bibdata":{"bibtype":"misc","type":"misc","abstract":"Multivariate time series forecasting has long received significant attention in real-world applications, such as energy consumption and traffic prediction. While recent methods demonstrate good forecasting abilities, they have three fundamental limitations. (i) Discrete neural architectures: Interlacing individually parameterized spatial and temporal blocks to encode rich underlying patterns leads to discontinuous latent state trajectories and higher forecasting numerical errors. (ii) High complexity: Discrete approaches complicate models with dedicated designs and redundant parameters, leading to higher computational and memory overheads. (iii) Reliance on graph priors: Relying on predefined static graph structures limits their effectiveness and practicability in real-world applications. In this paper, we address all the above limitations by proposing a continuous model to forecast $\\textbf{M}$ultivariate $\\textbf{T}$ime series with dynamic $\\textbf{G}$raph neural $\\textbf{O}$rdinary $\\textbf{D}$ifferential $\\textbf{E}$quations ($\\texttt{MTGODE}$). Specifically, we first abstract multivariate time series into dynamic graphs with time-evolving node features and unknown graph structures. Then, we design and solve a neural ODE to complement missing graph topologies and unify both spatial and temporal message passing, allowing deeper graph propagation and fine-grained temporal information aggregation to characterize stable and precise latent spatial-temporal dynamics. Our experiments demonstrate the superiorities of $\\texttt{MTGODE}$ from various perspectives on five time series benchmark datasets.","added-at":"2023-04-12T09:42:14.000+0200","author":[{"propositions":[],"lastnames":["Jin"],"firstnames":["Ming"],"suffixes":[]},{"propositions":[],"lastnames":["Zheng"],"firstnames":["Yu"],"suffixes":[]},{"propositions":[],"lastnames":["Li"],"firstnames":["Yuan-Fang"],"suffixes":[]},{"propositions":[],"lastnames":["Chen"],"firstnames":["Siheng"],"suffixes":[]},{"propositions":[],"lastnames":["Yang"],"firstnames":["Bin"],"suffixes":[]},{"propositions":[],"lastnames":["Pan"],"firstnames":["Shirui"],"suffixes":[]}],"biburl":"https://www.bibsonomy.org/bibtex/223ded3d3d60124236732b3deb0609001/manli","description":"Multivariate Time Series Forecasting with Dynamic Graph Neural ODEs","doi":"10.1109/TKDE.2022.3221989","howpublished":"https://github.com/GRAND-Lab/MTGODE","interhash":"62fd3e28a424edddc6265a90fe9b54a4","intrahash":"23ded3d3d60124236732b3deb0609001","keywords":"ma_sem2023","note":"cite arxiv:2202.08408Comment: 14 pages, 6 figures, 5 tables","timestamp":"2023-04-12T09:43:59.000+0200","title":"Multivariate Time Series Forecasting with Dynamic Graph Neural ODEs","url":"http://arxiv.org/abs/2202.08408","year":"2022","bibtex":"@misc{jin2022multivariate,\n abstract = {Multivariate time series forecasting has long received significant attention\r\nin real-world applications, such as energy consumption and traffic prediction.\r\nWhile recent methods demonstrate good forecasting abilities, they have three\r\nfundamental limitations. (i) Discrete neural architectures: Interlacing\r\nindividually parameterized spatial and temporal blocks to encode rich\r\nunderlying patterns leads to discontinuous latent state trajectories and higher\r\nforecasting numerical errors. (ii) High complexity: Discrete approaches\r\ncomplicate models with dedicated designs and redundant parameters, leading to\r\nhigher computational and memory overheads. (iii) Reliance on graph priors:\r\nRelying on predefined static graph structures limits their effectiveness and\r\npracticability in real-world applications. In this paper, we address all the\r\nabove limitations by proposing a continuous model to forecast\r\n$\\textbf{M}$ultivariate $\\textbf{T}$ime series with dynamic $\\textbf{G}$raph\r\nneural $\\textbf{O}$rdinary $\\textbf{D}$ifferential $\\textbf{E}$quations\r\n($\\texttt{MTGODE}$). Specifically, we first abstract multivariate time series\r\ninto dynamic graphs with time-evolving node features and unknown graph\r\nstructures. Then, we design and solve a neural ODE to complement missing graph\r\ntopologies and unify both spatial and temporal message passing, allowing deeper\r\ngraph propagation and fine-grained temporal information aggregation to\r\ncharacterize stable and precise latent spatial-temporal dynamics. Our\r\nexperiments demonstrate the superiorities of $\\texttt{MTGODE}$ from various\r\nperspectives on five time series benchmark datasets.},\n added-at = {2023-04-12T09:42:14.000+0200},\n author = {Jin, Ming and Zheng, Yu and Li, Yuan-Fang and Chen, Siheng and Yang, Bin and Pan, Shirui},\n biburl = {https://www.bibsonomy.org/bibtex/223ded3d3d60124236732b3deb0609001/manli},\n description = {Multivariate Time Series Forecasting with Dynamic Graph Neural ODEs},\n doi = {10.1109/TKDE.2022.3221989},\n howpublished = {https://github.com/GRAND-Lab/MTGODE},\n interhash = {62fd3e28a424edddc6265a90fe9b54a4},\n intrahash = {23ded3d3d60124236732b3deb0609001},\n keywords = {ma_sem2023},\n note = {cite arxiv:2202.08408Comment: 14 pages, 6 figures, 5 tables},\n timestamp = {2023-04-12T09:43:59.000+0200},\n title = {Multivariate Time Series Forecasting with Dynamic Graph Neural ODEs},\n url = {http://arxiv.org/abs/2202.08408},\n year = 2022\n}\n\n","author_short":["Jin, M.","Zheng, Y.","Li, Y.","Chen, S.","Yang, B.","Pan, S."],"key":"jin2022multivariate","id":"jin2022multivariate","bibbaseid":"jin-zheng-li-chen-yang-pan-multivariatetimeseriesforecastingwithdynamicgraphneuralodes-2022","role":"author","urls":{"Paper":"http://arxiv.org/abs/2202.08408"},"keyword":["ma_sem2023"],"metadata":{"authorlinks":{}}},"bibtype":"misc","biburl":"http://www.bibsonomy.org/bib/author/pan, shirui?items=1000","dataSources":["mKA5vx6kcS6ikoYhW","fcdT59YHNhp9Euu5k","m7B7iLMuqoXuENyof","AoeZNpAr9D2ciGMwa"],"keywords":["ma_sem2023"],"search_terms":["multivariate","time","series","forecasting","dynamic","graph","neural","odes","jin","zheng","li","chen","yang","pan"],"title":"Multivariate Time Series Forecasting with Dynamic Graph Neural ODEs","year":2022}