Working with pytorch to build different deep generative models.
- To introduce myself with pytorch. I started by building models of some Deep Learning models, as a basic fully connected feed-forward netowrk and a CNN.
- In this case, using a GRU and a cropus of Shakespear's text. I builded a model to learn how to replicate it.
- Here I explore different VAE architectures for digit generation with the MNIST dataset. The models I tried are:
- VAE with continuous latent spaces, modeling a Gaussian distribution.
- Beta-VAE, adding beta parameter to the VAE to control the desintanglement of the latent space.
- VAE with discrete latent spaces, 10 classes, one for each digit.