Hands-On Machine Learning for Algorithmic Trading

Conditional GANs

cGANs introduce additional label information into the training process, resulting in better quality and some control over the output.

cGANs alter the baseline architecture displayed precedingly by adding a third input value to the discriminator, which contains class labels. These labels, for example, could convey gender or hair color information when generating images.

Extensions include the Generative Adversarial What-Where Network (GAWWN), which uses bounding-box information to not only generate synthetic images but also place objects at a given location.