Learn One Size to Infer All: Exploiting Symmetries in Dynamical Systems Using Scalable Neural Networks

Explore innovative approaches to inferring dynamical systems using scalable neural networks in this 56-minute lecture from the Instituto de Física Interdisciplinar y Sistemas Complejos (IFISC). Delve into the concept of exploiting symmetries to enhance learning efficiency across various system sizes. Examine key topics including physics-informed learning, delay systems, delayed echo state networks, and the reservoir computing framework. Gain insights into accuracy considerations, attractor selection, and special temporal systems. Discover how these advanced techniques can be applied to understand and predict complex dynamical behaviors in interdisciplinary fields of physics and complex systems.

Introduction
Dynamical Systems
Outline
Parading
Physics Informed Learning
Delay Systems
Delayed EchoState Network
Accuracy
Which attractor to learn
Special temporal systems
Reservoir Computing Framework
Conclusion