AI Paper Atlas

Research map / Time series / Long-term forecasting

Time-Series Decomposition as a Standalone Task: A Mechanism-Driven Diagnostic Benchmark

Zipeng Wu, Jiani Wei, Shiqiao Zhou, Jiajun chen, Fabian Spill, J. W. Andrews

ICML 2026 · Time series › Long-term forecasting · approach: Attention for time series forecasting

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
decompose with STL, CEEMDAN or VMD before symbolic regressionPySR · Mauna Loa CO2Seasonal R2: 0.000 -> 0.897 (STL), 0.989 (VMD)Improved
decompose with SSA before ND2 symbolic regressionND2 · Linear-plus-sine syntheticReconstruction R2: near-zero -> R2 about 0.31Improved
decompose with STL before ND2 on multi-harmonic signalND2 · Linear-plus-multi-harmonic syntheticTrend SR R2: about 0.55 -> about 0.91Improved
decompose with STL before ND2 on multi-harmonic signalND2 · Linear-plus-multi-harmonic syntheticSeasonal SR R2: seasonal SR remains low (about 0.03)No clear effect
decompose before PySR when M2/S2 frequencies are closePySR · Tidal datasetSeasonal R2: seasonal R2 to the 0.46-0.60 rangeMixed
use FFT-estimated period instead of true period (ablation)STL · Period robustnessSeasonal max-lag correlation: 0.889 -> 0.798Hurt
misspecify the period by +20% (ablation)STL · Period robustnessSeasonal max-lag correlation: 0.889 -> 0.405Hurt
misspecify the period by +20% (ablation)MSTL · Period robustnessSeasonal max-lag correlation: 0.913 -> 0.338Hurt
use unsupervised peak matching instead of injected-period matching (ablation)CEEMDAN · Alignment robustnessAggregate score: 0.727 -> 0.355Hurt
use truth-guided component matching instead of injected matching (ablation)VMD · Alignment robustnessAggregate score: -0.283 -> 0.559Improved
bounded common-budget tuning of VMD (ablation)VMD · Robustness subset (5 scenarios)Aggregate rank order: STL > Wavelet > VMD -> VMD > STL > WaveletImproved
frequency drift scenario: fixed-period prior (STL) vs SSASTL vs SSA · Frequency-drift scenarioSeasonal max-lag correlation: 0.967 (SSA) vs 0.790 (STL)Hurt

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