anjali88/vae-mxnet

MXNet/Gluon implementation of the original (Gaussian) Variational Autoencoders (VAE)

Variational autoencoders in MXNet/Gluon

Implementations of variational autoencoders using MXNet/Gluon.

References:

1. Kingma, Diederik P., and Max Welling. "Auto-encoding variational bayes." arXiv preprint arXiv:1312.6114 (2013).
2. Rezende, Danilo Jimenez, Shakir Mohamed, and Daan Wierstra. "Stochastic backpropagation and approximate inference in deep generative models." arXiv preprint arXiv:1401.4082 (2014).

Code:

• Implementation using MXNet API (i.e. mxnet.sym, mxnet.mod) vae-mxnet.ipynb
• Implementation using Gluon API (i.e. gluon.HybridBlock, autograd) vae-gluon.ipynb
• CNN-based version, implemented using Gluon vaecnn-gluon.ipynb

Results:

• Generated MNIST figures by randomly sampling learned latent space

With 2-D latent space	With 10-D latent space	With 20-D latent space

• Learned 2-D manifold

Latent feature Z corresponding to 1000 test images	Generated images from grid scan in Z