To address these issues, we propose a self-supervised intrinsic rewardELIGN expectation alignment - inspired by the self-organization principle in Zoology.

Next, we prove rigorous bounds on the cumulative group pseudoregret obtained byFEDUCB.

Automated scientific discovery promises to accelerate progress across scientific domains. However, developing and evaluating an AI agent's capacity for endto-end scientific reasoning is challenging as running real-world experiments is often prohibitively expensive or infeasible.