Show HN: Koala: Cross-Substrate Tactical Representations in Graph-Based Go [pdf]

Category: ai-ml

Tags: go, graph-neural-network, monte-carlo-tree-search

Score: 6.3/10 (Innovation: 8, Technical: 9, Documentation: 4, Utility: 4)

Koala is a research project exploring a coordinate-free, topology-aware graph neural network combined with Monte Carlo Tree Search to play Go on diverse Euclidean tilings and 3-D lattices, demonstrating zero-shot transfer to unseen board geometries. Its innovative combination of GNNs, MCTS, and cross-substrate transfer pushes beyond traditional Go AI, though it remains an academic paper with limited practical utility and minimal documentation beyond the paper itself.

Target audience: data engineers, ai researchers

Repository: https://github.com/vonduffen/koala-paper/blob/main/koala.pdf · CSS

View on Hacker News