Generative Models Training by Markus Liedl
Generative Models capture the essence of untagged data. They prove that fact by their ability to generate new data that looks original. In this course we'll look at many proposed models:
- DCGAN (deep convolutional GAN)
- EBGAN (engery based GAN)
- ALI (a GAN featuring decoder and encoder)
- Wasserstein GAN
- RPGAN (multiple discriminators after random projections)
- GLO (generative latent optimization; kind of an auto-encoder without an encoder)
The main work with generative models is to find hyper parameters that perform well. Judging the quality of a trained model is difficult to automate.
There are open data sets and I bring some data as well. Your are invited to bring your own data. (images)
If you are interested in the course mail me at firstname.lastname@example.org. Normally the course will happen in or near Munich, Germany. I'm also offering to join your team and work together with you on specific topics.
This course is available in English and German.
Related Deep Learning courses are