Install OpenAI gym on Windows 10
I am very much interested in the world models approach to reinforcement learning.
- Research paper World models by D. Ha and J. Schmidhuber
- Interactive paper Recurrent world models facilitate policy evaluation
- Reimplementation of part of the world models paper
The world models approach uses the following ideas:
- Variational autoencoders
- Recurrent Neural Netoworks
- Mixture Density Networks
Before diving into world models, let us review the above ideas. The review of VAE is based on chapter 3 Variational autoencoder in of the book Generative deep learning by D. Foster. I have put the code in the folder GDL_code.
TO DO:
Recurrent Neural Networks
Mixture Density Networks
Check if using the independent component layer from the recent paper Rethinking the Usage of Batch Normalization and Dropout leads to better performing variational autoencoders.