Stoofvlees II is the World Computer Chess Champion of 2023 and shared the title with Jonny and Raptor in 2024. It is a hybrid Monte–Carlo Tree-Search chess engine augmented by a deep-convolutional neural network and a classical evaluator. This paper reviews the architecture and training methodology. Moreover, it documents Stoofvlees II's performances at the 2023 (Valencia) and 2024 (Santiago de Compostela) WCCCs. Particular emphasis is placed on (1) the design decisions that deviate from the AlphaZero paradigm, (2) the training and hardware regimes employed, and (3) the practical tournament considerations that decided critical games. All three issues together were important for the engine's first place at the 2023 and 2024 WCCCs.
ChetlurS.WoolleyC.VandermerschP.CohenJ.TranJ.CatanzaroB.ShelhamerE. (2014). cuDNN: Efficient primitives for deep learning. arXiv preprint arXiv:1410.0759, https://arxiv.org/abs/1410.0759ar
2.
ClarkC.StorkeyA. (2015). Training deep convolutional neural networks to play go. In Proceedings of the 32nd international conference on machine learning (ICML) (Vol. 37, pp. 1766–1774). Proceedings of Machine Learning Research (PMLR).
CoulomR. (2006). Efficient selectivity and backup operators in Monte-Carlo tree search. In van den HerikH. J.CiancariniP.DonkersH. H. L. M. (Eds.), Computers and games. CG 2006. Lecture notes in computer science (Vol. 4630, pp. 72–83). Springer.
5.
DanihelkaI.GuezA.SchrittwieserJ.SilverD. (2022). Policy improvement by planning with Gumbel. In: Proceedings of the 10th international conference on learning representations (ICLR). https://openreview.net/forum?id=bERaNdoegnOOpenReview
6.
GellyS.WangY. (2007). Exploration–exploitation in Go: UCT for Monte Carlo Go. In SchölkopfB.PlattJ.HoffmanT. (Eds.), Advances in neural information processing systems 19 (NIPS 2006) (pp. 1009–1016). MIT Press.
7.
HuJ.ShenL.SunG. (2018). Squeeze-and-excitation networks. In Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR) (pp. 7132–7141), https://doi.org/10.1109/CVPR.2018.00745
8.
KidambiR.NetrapalliP.JainP.KakadeS. (2018). On the insufficiency of existing momentum schemes for stochastic optimization. In: Proceedings of the international conference on learning representations (ICLR). https://openreview.net/forum?id=ryQu7f-RZ
NasuY. (Apr. 2018). Efficiently updatable neural-network-based evaluation function for computer Shogi. Presented at the 28th World Computer Shogi Championship. https://github.com/asdfjkl/nnue
PijlR.PascuttoG.van den HerikJ. (2024a). The 2024 world chess software championship (WCSC) in Santiago de compostela, Spain on October 19–20, 2024. ICGA Journal, 46(3–4), 132–140. https://doi.org/10.1177/13896911251320745
16.
PijlR.PascuttoG.van den HerikJ. (2024b). The 2024 world computer chess championship (WCCC) in Santiago de Compostela, Spain on October 21–23, 2024. ICGA Journal, 46(3–4), 127–131. https://doi.org/10.1177/13896911251320803
17.
SilverD.HuangA.MaddisonC. J.GuezA.SifreL.van den DriesscheG.SchrittwieserJ.AntonoglouI.PanneershelvamV.LanctotM.DielemanS.GreweD.NhamJ.KalchbrennerN.SutskeverI.LillicrapT.LeachM.KavukcuogluK.GraepelT.HassabisD. (2016). Mastering the game of Go with deep neural networks and tree search. Nature, 529(7587), 484–489, https://doi.org/10.1038/nature16961
18.
SilverD.HubertT.SchrittwieserJ.AntonoglouI.LaiM.GuezA.LanctotM.SifreL.KumaranD.GraepelT.LillicrapT. P.SimonyanK.HassabisD. (Dec. 2018). A general reinforcement learning algorithm that masters Chess, Shogi, and Go through self-play. Science, 362(6419), 1140–1144, https://doi.org/10.1126/science.aar6404
