SSNet: Skip and Split MLP Network for Long-Term Series Forecasting

Pengfei Tian, Erli Meng, Tianzhe Dong, Teng Wu, Yuan Chen

Keywords: Time Series Forecasting, Deep Learning, MLP

Abstract:

Time series forecasting is critical across various domains,
including energy, transportation, weather prediction, and healthcare.
Although recent advances using CNNs, RNNs, and Transformer-based models have shown promise,
these approaches often suffer from architectural complexity and low computational efficiency.
MLP-based networks offer better computational efficiency,
and some frequency-domain MLP models have demonstrated the ability to handle periodic time series data.
However, standard MLP-based methods still struggle to directly model periodic and temporal dependencies in the time domain,
which are essential for accurate time series forecasting.
To address these challenges, we propose the Skip and Split MLP Network (SSNet),
featuring innovative Skip-MLP and Split-MLP components that enable MLP models to directly capture periodicity and temporal dependencies in the time domain.
SSNet requires fewer parameters than traditional MLP-based architectures,
improving computational efficiency.
Empirical results on multiple real-world long-term forecasting datasets demonstrate that SSNet significantly outperforms state-of-the-art models,
delivering better performance with fewer parameters. Notably, even a single Skip-MLP unit matches the performance of high-performing models like PatchTST.

Primary Area: learning on time series and dynamical systems

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 4308