AI Paper Atlas

Research map / Time series / Long-term forecasting

From Fourier to Koopman: Spectral Methods for Long-term Time Series Prediction

Henning Lange, Steven L. Brunton, J. Nathan Kutz

JMLR 2021 · Time series › Long-term forecasting · approach: Lightweight long-term forecasters · 69 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use Koopman (nonlinear decoder) instead of Fourier basisspectral forecaster · RE-Europe energy demand (horizon 100%)relative cumulative error: 0.3 -> 0.19Improved
use spectral Koopman forecaster instead of LSTMlong-term forecaster · RE-Europe energy demand (horizon 100%)relative cumulative error: 0.45 -> 0.19Improved
replace one-step recursive GRU with global Fourier fitlong-term forecaster · RE-Europe energy demand (horizon 100%)relative cumulative error: 0.5 -> 0.3Improved
set CW-RNN internal clocks to periods found in the data (ablation)CW-RNN · RE-Europe energy demand (horizon 100%)relative cumulative error: 1.09 -> 1.15No clear effect
replace PCA+Fourier with Koopman and transpose-conv decoderspectral forecaster · Traveling wave (synthetic)reconstruction error: 6% -> 1%Improved
use Fourier algorithm instead of DMDPCA-reduced flow forecaster · Kolmogorov 2D (Re~10) (horizon 100%)relative cumulative error: 165% -> 1.69%Improved
use Fourier algorithm instead of DMDPCA-reduced flow forecaster · Flow over cavity (Re 7500) (horizon 100%)relative cumulative error: 63.8% -> 1%Improved
use Fourier algorithm instead of DMDPCA-reduced flow forecaster · Cylinder flow (Re 100) (horizon 100%)relative cumulative error: 4.69% -> 0.1%Improved
use Fourier algorithm instead of DMDPCA-reduced flow forecaster · Video frame prediction (horizon 100%)relative cumulative error: 22% -> 27.3%Hurt
increase training data from 2000 to 8000 samplesFourier and Koopman forecasters · synthetic nonlinear oscillatorrelative cumulative errorImproved
increase measurement noise varianceFourier and Koopman forecasters · synthetic nonlinear oscillator (T=4000)relative cumulative errorHurt
use Koopman instead of Fourierspectral forecaster · synthetic nonlinear oscillatorrelative cumulative errorImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy