Adaptive Learning for Learning World Models

So I joined the Bayesian lab. So I was thinking what expertise do we have here that can be applied to these models that I care about, but they're in a quite different area of research wherever else used to work at. We worked on was this idea of active learning for learning world models. What you want is that data to be useful to improve the world model rather than just being kind of like your greedy policy that you just deployed, add a bit of noise to it and hope it's good. What you probably want to do is also have some objective, but explicitly seeks out states that improves the world model. And then basically what happened was offline RL just became a big thing