Enhancing Time-Series Forecasting with Iterative Decomposition and Separable Training
Keywords: time series forecasting, training method
Abstract: Time series data, crucial for decision-making in fields like finance and healthcare, often presents challenges due to its inherent complexity, exacerbating the bias-variance tradeoff and leading to overfitting and underfitting in conventional forecasting models. While promising, state-of-the-art models like PatchTST, iTransformer, and DLinear are hindered by this tradeoff, limiting their ability to separate predictable patterns from noise. To resolve this, we propose the IDEAS framework, which reduces the bias-variance tradeoff to help models achieve optimal performance. IDEAS combines iterative residual decomposition, which reduces bias by extracting predictable patterns, and separable training, which reduces variance by independently optimizing each component. We provide theoretical proof and demonstrate through experiments that IDEAS significantly improves performance across four state-of-the-art models on nine complex benchmark datasets, offering a more robust solution for complex time series forecasting.
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.
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: 1408
Loading