Suggestions for further work (possibly outside the scope of this course):
- Use a variational auto-encoder (VAE)
- Consider using the embeddings as the index to the Q-table directly
- Aggregate frames (optical flow)
- Do not train the AE sequentially
- Implementation
- Final paper - Please submit it to EasyChair
- Final presentation slides