Keep pulling the thread on Noam Brown.
The Update-Equivalence Framework for decision-time planning is based on update equivalence rather than on solving subgames.
A provably sound search algorithm for fully cooperative games was derived based on mirror descent using the Update-Equivalence Framework.
A search algorithm for adversarial games was derived based on magnetic mirror descent using the Update-Equivalence Framework.
In the game Hanabi, a search algorithm based on mirror descent exceeds or matches the performance of public information-based search methods while using two orders of magnitude less search time.
The mirror descent algorithm's performance in Hanabi is the first instance of a non-public-information-based algorithm outperforming public-information-based approaches in a domain they have historically dominated.
The Update-Equivalence Framework enables scalability to games with large amounts of non-public information because its algorithms replicate the updates of last-iterate algorithms, which do not rely on public information.