HyperGAN 0.10 preview 1

GANs are a powerful form of unsupervised learning. HyperGAN aims to bring this exciting form of AI to everyone. All you need is a graphics card.

128x128 trained at batch size 1 for 2 million steps.
Batch walking the above model(linear interpolation along latent).

HyperGAN is a community project. Everyone is welcome to use and contribute! See the github and the active discord to find out more. https://github.com/HyperGAN/HyperGAN and https://discord.gg/t4WWBPF

New model authoring domain specific language(DSL):

Model authors can now create amazing model architectures completely in JSON! No python coding required.

Online Learning:

Online learning is the ability to learn new tasks while not forgetting old tasks. Integrating these features enables training at batch size 1, as well as general improvements to stability.

We introduce several features to help enable online learning:

And more!

0.10 is currently in open alpha. Pip has not yet been updated, you must install from source. See the discord for 0.10 installation instructions and configurations.

The developers also offer consulting services. Contact us today to find out how we can help your product or business: hypergan@protonmail.com