The talk covers generative learning, a branch of AI that learns to generate new samples from unknown distributions based on existing data. Gottschalk discusses modern generative learning challenges, notably the high dimensionality of data such as images or text, and the use of Generative Adversarial Networks (GANs) to model this data.

A significant part of the presentation is dedicated to the application of GANs in fluid dynamics, specifically in simulating turbulence, a chaotic and computationally intensive process. The team demonstrates that GANs can effectively model turbulence, producing results comparable to large eddy simulations (LES) but with less computational effort. This is further explored through practical examples like turbulence modelling in video game engines and real-world scenarios like the Karman Vortex Street and LPT Stator.

The talk also delves into the theoretical aspects of generative learning, discussing concepts like ergodicity, which relates to the statistical properties of chaotic systems. The success of GANs in learning these properties from data is highlighted, along with their application in inverse design through Invertible Neural Networks (INN). The presentation concludes by emphasising the utility of generative learning in various fields, its potential in combining with different data models like deterministic chaos, and the significant speedup it offers during inference.