2019-09-14, 14:30–15:00, Assembly Room
A beginner-friendly introduction to how GANs work, and how you can write your own with PyTorch. Artists, designers and creative coders very welcome!
Generative Adversarial Networks (GANs) are a hot topic in AI research right now. As recently as 2018, the art world saw its first AI generated portrait auctioned for $432,500. Today we see tech giants showcasing their prowess producing AI-generated images that are unsettlingly realistic.
The aim of this talk is to demystify machine learning, how neural networks work, and the uniquely clever adversarial training method, popularised as recently as 2014 by Ian Goodfellow, considered by many as inventor of GANs.
We'll also see how easy it is to implement our ideas in PyTorch, including creating a GAN in only a few lines of code.
The talk is suitable for beginners and very open to artists and designers. At minimum you'll gain an intuition into machine learning and adversarial training, and the confidence to code or try GANs yourself.