The Very Bad Transcription of Uncertainty in Task Inference

Very bad treatment of uncertainty is so cool that makes it really special to me. The magic of variational inference plus conditioning on that uncertainty for the meta learning allows you to learn actually optimal exploration pretty easily. There's also like unsupervised learning which could be useful if you just don't have access to rewards at test time. Some algorithms that do this are like heavy and learning algorithms there's a lot of research into them. They're local and they're unsupervised in their inner loop but the outer loop as we mentioned still uses rewards.