Efficient Time Series Forecasting via Hyper-Complex Models and Frequency Aggregation

EYAL YAKIR, Dor Tsur, Haim H. Permuter

Keywords: time-series forecasting, frequency models, hyper-complex machine learning, short-time Fourier transform

TL;DR: We propose a novel time-series forecasting model that uses STFT window aggregation and hyper-complex models.

Abstract:

Time-series forecasting is a long-standing challenge in statistics and machine learning, with one of the key difficulties being the ability to process sequences with long-range dependencies. A recent line of work has addressed this by applying the short-time Fourier transform (STFT), which partitions sequences into multiple subsequences and applies a Fourier transform to each separately.
We propose the Frequency Information Aggregation (FIA-Net), a model that can utilize two backbone architectures: the Window-Mixing MLP (WM-MLP), which aggregates adjacent window information in the frequency domain, and the Hyper-Complex MLP (HC-MLP), which treats the set of STFT windows as hyper-complex (HC) valued vectors. and employ HC algebra to efficiently combine information from all STFT windows altogether. Furthermore, due to the nature of HC operations, the HC-MLP uses up to three times fewer parameters than the equivalent standard window aggre- gation method. We evaluate the FIA-Net on various time-series benchmarks and show that the proposed methodologies outperform existing state-of-the-art meth- ods in terms of both accuracy and efficiency. Our code is publicly available on https://anonymous.4open.science/r/research-1803/

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: 12311