Residual Integration Neural Network

Said Ouala (1) , Ananda Pascual (2) , Ronan Fablet (1, 3)

1 IMT Atlantique - SC - Département Signal et Communications
2 IMEDEA - Institut Mediterrani d'Estudis Avancats
3 Lab-STICC_IMTA_CID_TOMS

In this work, we investigate residual neural network representations for the identification and forecasting of dynamical systems. We propose a novel architecture that jointly learns the dynamical model and the associated Runge-Kutta integration scheme. We demonstrate the relevance of the proposed architecture with respect to learning-based state-of-the-art approaches in the identification and forecasting of chaotic dynamics when provided with training data with low temporal sampling rates.

Mots clés: Dynamical systems; Data-driven models; Neural networks; Forecasting

Domaines: Machine Learning [stat.ML]

Said Ouala, Ananda Pascual, Ronan Fablet. Residual Integration Neural Network. ICASSP 2019 : IEEE International Conference on Acoustics, Speech and Signal Processing, May 2019, Brighton, United Kingdom. ⟨10.1109/ICASSP.2019.8683447⟩. ⟨hal-02005399⟩