Sonnet: Spectral Operator Neural Network for Multivariable Time Series Forecasting

Yuxuan Shu, Vasileios Lampos · 2025

Read Sonnet: Spectral Operator Neural Network for Multivariable Time Series Forecasting by Yuxuan Shu, Vasileios Lampos (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-15312.txt.

Introduction

Yuxuan Shu and Vasileios Lampos present Sonnet, a spectral network for forecasting many linked series in this 2025 paper. Extra input series can sharpen forecasts when they truly relate. Plain transformers often miss those cross-variable ties over time.

Sonnet first reshapes inputs with learnable wavelet transforms. A Koopman operator lens then reads the series in spectral terms. Its key part, coherence attention, links variables by how their rhythms align. Across dozens of tasks the model won most head-to-head comparisons. Average error fell a few points against the strongest rival. Plugging its attention into other models cut their errors too, most on the hardest tasks.

Catalog record

CREATORYuxuan Shu, Vasileios Lampos
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE131 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-15312

Browse the full library · About the Tower of Babel