The world is being quietly rearranged by people who write very long documents.


The title they went with Extreme Value Monte Carlo Tree Search for Classical Planning Noisy translates that to

Better algorithm for planning problems that don't fit old math


Researchers found that planning algorithms borrowed from board games use the wrong mathematical assumptions for certain types of problems — specifically, they assume rewards are bounded when they're actually unbounded. They developed a new algorithm based on extreme value theory that better handles these open-ended cost estimates and proved it works mathematically.
This is a specialist paper that improves how computers solve planning problems, but only within research settings — there's no evidence this algorithm is being used to solve real-world planning at scale or that it changes what's economically or operationally feasible in any deployed system.

If you insist
Read the original →