Promoting Data-Efficiency of Deep Learning for Dynamical Systems

Explore a cutting-edge approach to deep learning for dynamical systems in this 56-minute talk from the Alan Turing Institute. Delve into the challenges of data scarcity in scientific research and discover how incorporating Euclidean symmetry can revolutionize neural network-based reduced-order models. Learn about the Euclidean symmetric neural network (ESNN) and its ability to generalize from limited data to various unseen conditions. Examine the application of ESNN to a 3D Slinky system, demonstrating energy invariance and force equivariance. Understand how this approach leads to significant simulation acceleration compared to traditional numerical methods. Gain insights into the components of the learning pipeline through an ablation study, highlighting the potential of this innovative technique for more efficient deep learning in dynamical systems.