Learning to Plan via Supervised Contrastive Learning and Strategic Interpolation: A Chess Case Study Conference: Undergraduate and Master’s Consortium at KDD 2025; August 03–07, 2025; Toronto, Canada ISBN: 978-1-4503-XXXX-X/25/08 CCS: Computing methodologies Knowledge representation and reasoning CC

Andrew Hamara, Greg Hamerly, Pablo Rivas, Andrew C. Freeman · 2025

Read Learning to Plan via Supervised Contrastive Learning and Strategic Interpolation: A Chess Case Study Conference: Undergraduate and Master’s Consortium at KDD 2025; August 03–07, 2025; Toronto, Canada ISBN: 978-1-4503-XXXX-X/25/08 CCS: Computing methodologies Knowledge representation and reasoning CC by Andrew Hamara, Greg Hamerly, Pablo Rivas, Andrew C. Freeman (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-04892.txt.

Description

Andrew Hamara and co-authors teach a model to plan chess moves the way human intuition works, in a 2025 paper. The catalog title adds conference metadata, so the head title governs. Engines win by deep search, while humans pick promising moves by feel and check them shallowly. The team trained a transformer to place board states in a space ordered by position quality.

In that space, distance mirrors how alike two positions score, and paths between states read clearly. Move choice then means stepping toward better regions, with only a shallow beam search on top. The model reached a strong grandmaster level rating estimate. Bigger models and wider spaces did better, hinting that planning in a learned space could rival search in other perfect information games too.

Catalog record

CREATORAndrew Hamara, Greg Hamerly, Pablo Rivas, Andrew C. Freeman
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE18 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-04892

Browse the full library · About the Tower of Babel