Communication Dans Un Congrès Année : 2025

Winner-takes-all for Multivariate Probabilistic Time Series Forecasting

Résumé

We introduce TimeMCL, a method leveraging the Multiple Choice Learning (MCL) paradigm to forecast multiple plausible time series futures. Our approach employs a neural network with multiple heads and utilizes the Winner-Takes-All (WTA) loss to promote diversity among predictions. MCL has recently gained attention due to its simplicity and ability to address ill-posed and ambiguous tasks. We propose an adaptation of this framework for time-series forecasting, presenting it as an efficient method to predict diverse futures, which we relate to its implicit quantization objective. We provide insights into our approach using synthetic data and evaluate it on real-world time series, demonstrating its promising performance at a light computational cost.

Fichier principal
Vignette du fichier
main_icml.pdf (8.32 Mo) Télécharger le fichier

Dates et versions

hal-05100125 , version 1 (05-06-2025)
hal-05100125 , version 2 (11-08-2025)

Licence

Identifiants

Citer

Adrien Cortés, Rémi Rehm, Victor Letzelter. Winner-takes-all for Multivariate Probabilistic Time Series Forecasting. ICML 2025 : The 42nd International Conference on Machine Learning, Jul 2025, Vancouver (CA), Canada. ⟨hal-05100125v2⟩
3660 Consultations
575 Téléchargements

Altmetric

Partager

  • More