This lecture uses a hierarchical neural network architecture to learn flow maps for physical processes occurring on different time scales. The methods are especially successful in learning how to advance solutions on long-time scales, thus circumventing numerical stiffness issues typically associated with fast scale physics effects. The success of the method is demonstrated on a number of problems.
![](https://i.ytimg.com/vi/QZumxTzix2Q/mqdefault.jpg)