Shared Knowledge Ensemble of Search Trees (SKEST) is a novel search algorithm proposed by Google DeepMind in February 2025. It can execute multiple beam searches with different configurations in parallel and allow them to help each other through a knowledge sharing mechanism. The relevant research results were published in the paper "Gold-medalist Performance in Solving Olympiad Geometry with AlphaGeometry2".

The working principle of SKEST is as follows:

In each search tree, a node corresponds to an auxiliary construction attempt followed by an attempt to run the symbolic engine. If the attempt succeeds, all search trees terminate. If the attempt fails, the node writes the facts that the symbolic engine managed to prove to a shared fact database. These shared facts are filtered so that they are not auxiliary points specific to the node itself, but only relevant to the original problem. This way, these facts can also help other nodes in the same search tree, as well as nodes in different search trees.