AI Paper Atlas

Research map / Time series / Long-term forecasting

LagLLM: LLM-empowered lead–lag dependency learning for spatial-temporal time series forecasting

Binqing Wu, Jian Zhou, Zongjiang Shang, Ling Chen

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
add the lead-lag graph (ablation)LagLLM · PEMS08MAE: 14.14 -> 13.21Improved
add the lead-lag graph (ablation)LagLLM · ChinaAQIMAE: 19.74 -> 19.14Improved
add the data-driven mask to the hybrid graph (ablation)LagLLM · PEMS08MAE: 13.58 -> 13.21Improved
add the knowledge-driven (LLM semantic) mask (ablation)LagLLM · PEMS08MAE: 13.63 -> 13.21Improved
add structural token sorting (ablation)LagLLM · PEMS08MAE: 13.34 -> 13.21Improved
sort tokens by lead-lag structure instead of spatial topology (ablation)LagLLM · PEMS08MAE: 13.28 -> 13.21Improved
add the grouping loss (ablation)LagLLM · PEMS08MAE: 13.24 -> 13.21Improved
add lead-lag criteria to the prompt (ablation)LagLLM · PEMS08MAE: 13.33 -> 13.21Improved
use hybrid template-plus-learnable prompt instead of fully learnable prompt (ablation)LagLLM · PEMS08MAE: 13.59 -> 13.21Improved
replace GPT-2 backbone with LLaMA-3-8B (ablation)LagLLM · PEMS08MAE: 13.21 -> 13.19Improved
LagLLM vs STD-PLM with 5% training dataSTD-PLM · PEMS07 (5% training data)MAE: 32.48 -> 30.07Improved
LagLLM vs STD-PLM with 10% training dataSTD-PLM · PEMS07 (10% training data)MAE: 26.37 -> 23.82Improved

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