Paper: Ha and Schmidhuber, "World Models", 2018. https://doi.org/10.5281/zenodo.1207631. This repository is based on the work of the authors listed below. It adds to their work the Doom simulation from the original paper, and also explores a new direction: Co-training the MDN-RNN and VAE from the original paper in order to encourage the VAE to encode more salient features. Results are reported in a paper to be released shortly.
- Corentin Tallec - ctallec
- Léonard Blier - leonardblier
- Diviyan Kalainathan - diviyan-kalainathan
This project is licensed under the MIT License - see the LICENSE.md file for details