SCINet: Time Series Modeling and Forecasting with Sample Convolution and
Interaction
M. Liu, A. Zeng, M. Chen, Z. Xu, Q. Lai, L. Ma, and Q. Xu. (2021)cite arxiv:2106.09305Comment: This paper presents a novel convolutional neural network for time series forecasting, achieving significant accuracy improvements.
Abstract
One unique property of time series is that the temporal relations are largely
preserved after downsampling into two sub-sequences. By taking advantage of
this property, we propose a novel neural network architecture that conducts
sample convolution and interaction for temporal modeling and forecasting, named
SCINet. Specifically, SCINet is a recursive downsample-convolve-interact
architecture. In each layer, we use multiple convolutional filters to extract
distinct yet valuable temporal features from the downsampled sub-sequences or
features. By combining these rich features aggregated from multiple
resolutions, SCINet effectively models time series with complex temporal
dynamics. Experimental results show that SCINet achieves significant
forecasting accuracy improvements over both existing convolutional models and
Transformer-based solutions across various real-world time series forecasting
datasets. Our codes and data are available at
https://github.com/cure-lab/SCINet.
Description
SCINet: Time Series Modeling and Forecasting with Sample Convolution and Interaction
cite arxiv:2106.09305Comment: This paper presents a novel convolutional neural network for time series forecasting, achieving significant accuracy improvements
%0 Generic
%1 liu2021scinet
%A Liu, Minhao
%A Zeng, Ailing
%A Chen, Muxi
%A Xu, Zhijian
%A Lai, Qiuxia
%A Ma, Lingna
%A Xu, Qiang
%D 2021
%K time-series
%T SCINet: Time Series Modeling and Forecasting with Sample Convolution and
Interaction
%U http://arxiv.org/abs/2106.09305
%X One unique property of time series is that the temporal relations are largely
preserved after downsampling into two sub-sequences. By taking advantage of
this property, we propose a novel neural network architecture that conducts
sample convolution and interaction for temporal modeling and forecasting, named
SCINet. Specifically, SCINet is a recursive downsample-convolve-interact
architecture. In each layer, we use multiple convolutional filters to extract
distinct yet valuable temporal features from the downsampled sub-sequences or
features. By combining these rich features aggregated from multiple
resolutions, SCINet effectively models time series with complex temporal
dynamics. Experimental results show that SCINet achieves significant
forecasting accuracy improvements over both existing convolutional models and
Transformer-based solutions across various real-world time series forecasting
datasets. Our codes and data are available at
https://github.com/cure-lab/SCINet.
@misc{liu2021scinet,
abstract = {One unique property of time series is that the temporal relations are largely
preserved after downsampling into two sub-sequences. By taking advantage of
this property, we propose a novel neural network architecture that conducts
sample convolution and interaction for temporal modeling and forecasting, named
SCINet. Specifically, SCINet is a recursive downsample-convolve-interact
architecture. In each layer, we use multiple convolutional filters to extract
distinct yet valuable temporal features from the downsampled sub-sequences or
features. By combining these rich features aggregated from multiple
resolutions, SCINet effectively models time series with complex temporal
dynamics. Experimental results show that SCINet achieves significant
forecasting accuracy improvements over both existing convolutional models and
Transformer-based solutions across various real-world time series forecasting
datasets. Our codes and data are available at
https://github.com/cure-lab/SCINet.},
added-at = {2022-11-18T09:18:52.000+0100},
author = {Liu, Minhao and Zeng, Ailing and Chen, Muxi and Xu, Zhijian and Lai, Qiuxia and Ma, Lingna and Xu, Qiang},
biburl = {https://www.bibsonomy.org/bibtex/2e49069d3f31b81e80c5a9e8152e0e15e/manli},
description = {SCINet: Time Series Modeling and Forecasting with Sample Convolution and Interaction},
interhash = {b6488f253c04a922fa6d8b0985bbd7ec},
intrahash = {e49069d3f31b81e80c5a9e8152e0e15e},
keywords = {time-series},
note = {cite arxiv:2106.09305Comment: This paper presents a novel convolutional neural network for time series forecasting, achieving significant accuracy improvements},
timestamp = {2022-11-18T09:18:52.000+0100},
title = {SCINet: Time Series Modeling and Forecasting with Sample Convolution and
Interaction},
url = {http://arxiv.org/abs/2106.09305},
year = 2021
}