TODO.md

Hold off to do for reviewers?

Here are ideas I like and want to add. Hopefully the reviewers will ask some of the questions I expect.

Other agents

1. Max entopy exploration
2. ELBO/count-based exploration
3. Extrinsic reward exploration

Curiosity cat?

1. Train a VAE->MLP classifier with
   • random batches
   • softmax E batches
   • argmax E batches
2. Finetune a pretained-SOA model using determined curiosity?
   • Can curiosity further improve the state of the art?

ImageInfoBandits

• Leaning w/ VAE

Foraging tasks

1. A grid-maze
2. A scent landscape
3. A patch world