Variational Autoencoder and Generative Adversarial Network
Published:
In this project we aim to implement variational autoencoder and generative adversarial network on MNIST dataset. We will first derive the VAE algorithm and then implement the training and inference algorithms for VAE. We further experiment the VAE model with different number of nodes in the bottleneck layer in the middle of encoder-decoder architecture. Next we have implemented Convolution Neural Network for encoder and Deconvolution neural network for decoder. We then derive the Generative Adversarial Network algorithm and then implemented the algorithm to train on MNIST data to generate the digits