Hands-On Machine Learning for Algorithmic Trading

How GAN architectures are evolving

Since publication in 2014, GANs have experienced an enormous amount of interest, as evidenced by Yann LeCun's quote in the introduction, and have triggered a flurry of research.

The bulk of this work has refined the original architecture to adapt it to different domains and tasks, and expand it to include additional information, creating a conditional GAN (cGAN). Additional research has focused on improving methods for the challenging training process that requires achieving a stable game-theoretic equilibrium between two networks, each of which can be tricky to train on its own. The GAN landscape has become more diverse than we can cover here, but you can find additional references to both surveys and individual ...