VAEs and GANs for MNIST(UvA 2019 Deep Learning course Assignment)
This repository contains PyTorch implementations of Variational Auto Encoders and Generative Adversarial Networks and a report comparing their results and attributes based on MNIST data.
General questions regarding the topic and the models can be found in the assignment PDF. Insights and results can be found in the report.
Evenly sampled VAE results across 40 epochs, showing Bernoulli sample (left) and means (right). Latent space is 20 dimensional.
Evenly sampled GAN results across 150 epochs. Same sampling seed was used.
- Interpolation between digits
The code may bear resemblance to others, as these were assignments from the 2019 Deep Learning Course at the University of Amsterdam (UvA), as a part of Aritificial Intelligence MSc program. Template empty class codes were provided and we were tasked implement architectures, working training, and any other additional functionality we may have needed.