Abstract by McKell Stauffer
Variational Autoencoders and Reinforcement Learning
We research the possibilities of combining reinforcement and unsupervised deep learning methods. In particular, we are interested in creating a hierarchical combination of variational auto-encoders (VAEs), generative adversarial networks (GANs), and reinforcement learning (RL) methods. Additionally, we are interested in using a class-conditional VAE as a style transfer network. We will train our model on various Atari video game representations and hope to generalize the model to other Atari video games. Lastly, we are interested in expressing VAEs as a bilevel optimization frameworks.