Date:
Location:
Speaker: Petros Koumoutsakos, Gordon McKay Professor of Computing in Science and Engineering at Harvard University
Please contact Eleni Chatzi (chatzi@ibk.baug.ethz.ch) for details
Predictive simulations of complex systems are essential for applications ranging from weather forecasting to drug design. We present a novel systematic framework that bridges two of the key modalities for predicting complex system dynamics: large scale simulations and reduced order models. The framework of learned Effective Dynamics (LED) forms algorithmic alloys between machine learning algorithms and the equation-free approach for modeling complex systems.
The LED deploys autoencoders to formulate a mapping between fine and coarse grained representations and evolves the non-Markovian latent space dynamics using recurrent neural networks. The algorithm is validated on benchmark problems and we find that it outperforms state of the art reduced order models in terms of predictability and large scale simulations in terms of cost. The LED framework is applicable to systems ranging from chemistry to fluid mechanics and reduces the computational effort by up to five orders of magnitude while maintaining the prediction accuracy of the full system dynamics. We argue that LED provides a novel potent modality for the accurate prediction of complex systems.
For further information please visit http://www.zhcs.ch/news/events/distinguished-lecture-series/