TimeBase: The Power of Minimalism in Long-term Time Series Forecasting
Keywords: Long-term Time Series Forecasting
Abstract: Long-term time series forecasting (LTSF) has traditionally relied on models with large parameters to capture extended temporal dependencies. However, time series data, unlike high-dimensional images or text, often exhibit strong periodicity and low-rank structures, especially in long forecasting horizons. This characteristic can lead many models focusing on redundant patterns, resulting in inefficient use of computational resources. In this paper, we introduce TimeBase, an ultra-lightweight network with fewer than 0.4$k$ parameters, designed to harness the power of minimalism in LTSF. TimeBase extracts core periodic features by leveraging full-rank typical period representations under orthogonality constraints, enabling accurate prediction of future cycles. Extensive experiments on real-world datasets demonstrate that TimeBase not only achieves minimalism in both model size and computational cost, reducing MACs by 35x and parameter counts by over 1000 times compared to standard linear models, but also wins state-of-the-art forecasting performance, ranking Top1-Top5 in all 28 prediction settings. Additionally, TimeBase can also serve as a very effective plug-and-play tool for patch-based forecasting methods, enabling extreme complexity reduction without compromising prediction accuracy. Code is available at \url{https://anonymous.4open.science/r/TimeBase-fixbug}.
Primary Area: other topics in machine learning (i.e., none of the above)
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.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
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: 323
Loading