![[2501.16178] SWIFT: Mapping Sub-series with Wavelet Decomposition Improves Time Series Forecasting](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2501.16178] SWIFT: Mapping Sub-series with Wavelet Decomposition Improves Time Series Forecasting
Summary
The paper presents SWIFT, a lightweight model that enhances time series forecasting using wavelet decomposition, achieving state-of-the-art performance while being efficient for edge devices.
Why It Matters
As time series forecasting becomes increasingly vital in various applications, SWIFT addresses the challenge of deploying effective models in resource-constrained environments. This innovation could significantly improve forecasting accuracy and efficiency in real-world scenarios.
Key Takeaways
- SWIFT utilizes wavelet transform for efficient downsampling of time series data.
- The model achieves cross-band information fusion with a learnable filter.
- SWIFT's parameter count is significantly lower than traditional models, enhancing deployment feasibility.
- Comprehensive experiments demonstrate SWIFT's state-of-the-art performance across multiple datasets.
- The model is particularly suited for edge computing applications.
Computer Science > Machine Learning arXiv:2501.16178 (cs) [Submitted on 27 Jan 2025 (v1), last revised 14 Feb 2026 (this version, v3)] Title:SWIFT: Mapping Sub-series with Wavelet Decomposition Improves Time Series Forecasting Authors:Wenxuan Xie, Fanpu Cao View a PDF of the paper titled SWIFT: Mapping Sub-series with Wavelet Decomposition Improves Time Series Forecasting, by Wenxuan Xie and 1 other authors View PDF Abstract:In recent work on time-series prediction, Transformers and even large language models have garnered significant attention due to their strong capabilities in sequence modeling. However, in practical deployments, time-series prediction often requires operation in resource-constrained environments, such as edge devices, which are unable to handle the computational overhead of large models. To address such scenarios, some lightweight models have been proposed, but they exhibit poor performance on non-stationary sequences. In this paper, we propose $\textit{SWIFT}$, a lightweight model that is not only powerful, but also efficient in deployment and inference for Long-term Time Series Forecasting (LTSF). Our model is based on three key points: (i) Utilizing wavelet transform to perform lossless downsampling of time series. (ii) Achieving cross-band information fusion with a learnable filter. (iii) Using only one shared linear layer or one shallow MLP for sub-series' mapping. We conduct comprehensive experiments, and the results show that $\textit{SWIFT}$ ac...
