arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2106.09305cs.LGcs.AI
收藏

SCINet:样本卷积和交互作用的时间序列建模和预测

SCINet: Time Series Modeling and Forecasting with Sample Convolution and Interaction

The Chinese University of Hong Kong(香港中文大学)

Minhao Liu, Ailing Zeng, Muxi Chen, Zhijian Xu, Qiuxia Lai, Lingna Ma, Qiang Xu

摘要:

时间序列的一个独特性质是,在下采样为两个子序列之后,时间关系在很大程度上被保留。通过利用这一特性,我们提出了一种新的神经网络架构,进行样本卷积和交互的时间建模和预测,命名为SCINet。具体来说,SCINet是一个递归的下采样-卷积-交互架构。在每一层中,我们使用多个卷积滤波器从下采样的子序列或特征中提取独特而有价值的时间特征。通过结合这些从多个分辨率聚合的丰富特征,SCINet有效地对具有复杂时间动态的时间序列进行建模。实验结果表明,SCINet在各种真实世界时间序列预测数据集上,与现有的卷积模型和基于Transformer的解决方案相比,预测准确性得到了显着提高。我们的代码和数据可在https://github.com/cure-lab/SCINet上获得。

英文摘要:

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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑