Standard
A machine learning approach that beats Rubik's cubes. / Chervov, Alexander; Khoruzhii, Kirill; Bukhal, Nikita et al.
Advances in Neural Information Processing Systems: 38th Conference on Neural Information Processing Systems, NeurIPS 2025; Held 2-7 December 2025, San Diego, California, USA and 30 November - 5 December 2025, Mexico City, Mexico. Neural information processing systems foundation, 2025. p. 191182-191204 (Advances in Neural Information Processing Systems; Vol. 38).
Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Research › peer-review
Harvard
Chervov, A, Khoruzhii, K, Bukhal, N, Naghiyev, J, Zamkovoy, V, Koltsov, I, Cheldieva, L, Sychev, A, Lenin, A, Obozov, M, Urvanov, E & Romanov, AM 2025,
A machine learning approach that beats Rubik's cubes. in
Advances in Neural Information Processing Systems: 38th Conference on Neural Information Processing Systems, NeurIPS 2025; Held 2-7 December 2025, San Diego, California, USA and 30 November - 5 December 2025, Mexico City, Mexico. Advances in Neural Information Processing Systems, vol. 38, Neural information processing systems foundation, pp. 191182-191204, 38th Conference on Neural Information Processing Systems, San Diego, California, United States,
02.12.2025. <
https://www.proceedings.com/content/085/085713-5741open.pdf>
APA
Chervov, A., Khoruzhii, K., Bukhal, N., Naghiyev, J., Zamkovoy, V., Koltsov, I., Cheldieva, L., Sychev, A., Lenin, A., Obozov, M., Urvanov, E., & Romanov, A. M. (2025).
A machine learning approach that beats Rubik's cubes. In
Advances in Neural Information Processing Systems: 38th Conference on Neural Information Processing Systems, NeurIPS 2025; Held 2-7 December 2025, San Diego, California, USA and 30 November - 5 December 2025, Mexico City, Mexico (pp. 191182-191204). (Advances in Neural Information Processing Systems; Vol. 38). Neural information processing systems foundation.
https://www.proceedings.com/content/085/085713-5741open.pdfVancouver
Chervov A, Khoruzhii K, Bukhal N, Naghiyev J, Zamkovoy V, Koltsov I et al.
A machine learning approach that beats Rubik's cubes. In Advances in Neural Information Processing Systems: 38th Conference on Neural Information Processing Systems, NeurIPS 2025; Held 2-7 December 2025, San Diego, California, USA and 30 November - 5 December 2025, Mexico City, Mexico. Neural information processing systems foundation. 2025. p. 191182-191204. (Advances in Neural Information Processing Systems).
Author
Chervov, Alexander ; Khoruzhii, Kirill ; Bukhal, Nikita et al. /
A machine learning approach that beats Rubik's cubes. Advances in Neural Information Processing Systems: 38th Conference on Neural Information Processing Systems, NeurIPS 2025; Held 2-7 December 2025, San Diego, California, USA and 30 November - 5 December 2025, Mexico City, Mexico. Neural information processing systems foundation, 2025. pp. 191182-191204 (Advances in Neural Information Processing Systems).
BibTeX
@inproceedings{fdf4c4c4f8e344f5a572a8cda1ecafd8,
title = "A machine learning approach that beats Rubik's cubes",
abstract = "The paper proposes a novel machine learning-based approach to the pathfindingproblem on extremely large graphs. This method leverages diffusion distanceestimation via a neural network and uses beam search for pathfinding. We demonstrate its efficiency by finding solutions for 4x4x4 and 5x5x5 Rubik{\textquoteright}s cubes with unprecedentedly short solution lengths, outperforming all available solvers and introducing the first machine learning solver beyond the 3x3x3 case. In particular, it surpasses every single case of the combined best results in the Kaggle Santa 2023 challenge, which involved over 1,000 teams. For the 3x3x3 Rubik{\textquoteright}s cube, our approach achieves an optimality rate exceeding 98%, matching the performance of task-specific solvers and significantly outperforming prior solutions such as DeepCubeA (60.3%) and EfficientCube (69.6%). Our solution in its current implementation is approximately 25.6 times faster in solving 3x3x3 Rubik{\textquoteright}s cubes while requiring up to 8.5 times less model training time than the most efficient state-of-the-art competitor. Finally, it is demonstrated that even a single agent trained using a relatively small number of examples can robustly solve a broad range of puzzles represented by Cayley graphs of size up to 10145, confirming the generality of the proposed method.",
keywords = "pathfinding, multi-agent, ML, machine learning, Cayley graph, Rubik, cube, MLP",
author = "Alexander Chervov and Kirill Khoruzhii and Nikita Bukhal and Jalal Naghiyev and Vladislav Zamkovoy and Ivan Koltsov and Lyudmila Cheldieva and Arsenii Sychev and Arsenii Lenin and Mark Obozov and Egor Urvanov and Romanov, {Alexey M.}",
note = "We would like to thank MathWorks for their generous contribution of prizes for the BioMasster data competitions (more information available at https://www.drivendata.org/competitions/99/biomass-estimation/page/534/) and express their gratitude to all the participants. This research work is also part of the EO-AI4GlobalChange project funded by Digital Futures.; 38th Conference on Neural Information Processing Systems, NeurIPS 2025 ; Conference date: 02-12-2025 Through 07-12-2025",
year = "2025",
month = sep,
day = "19",
language = "English",
series = "Advances in Neural Information Processing Systems",
publisher = "Neural information processing systems foundation",
pages = "191182--191204",
booktitle = "Advances in Neural Information Processing Systems",
address = "United States",
}
RIS
TY - GEN
T1 - A machine learning approach that beats Rubik's cubes
AU - Chervov, Alexander
AU - Khoruzhii, Kirill
AU - Bukhal, Nikita
AU - Naghiyev, Jalal
AU - Zamkovoy, Vladislav
AU - Koltsov, Ivan
AU - Cheldieva, Lyudmila
AU - Sychev, Arsenii
AU - Lenin, Arsenii
AU - Obozov, Mark
AU - Urvanov, Egor
AU - Romanov, Alexey M.
N1 - Conference code: 38
PY - 2025/9/19
Y1 - 2025/9/19
N2 - The paper proposes a novel machine learning-based approach to the pathfindingproblem on extremely large graphs. This method leverages diffusion distanceestimation via a neural network and uses beam search for pathfinding. We demonstrate its efficiency by finding solutions for 4x4x4 and 5x5x5 Rubik’s cubes with unprecedentedly short solution lengths, outperforming all available solvers and introducing the first machine learning solver beyond the 3x3x3 case. In particular, it surpasses every single case of the combined best results in the Kaggle Santa 2023 challenge, which involved over 1,000 teams. For the 3x3x3 Rubik’s cube, our approach achieves an optimality rate exceeding 98%, matching the performance of task-specific solvers and significantly outperforming prior solutions such as DeepCubeA (60.3%) and EfficientCube (69.6%). Our solution in its current implementation is approximately 25.6 times faster in solving 3x3x3 Rubik’s cubes while requiring up to 8.5 times less model training time than the most efficient state-of-the-art competitor. Finally, it is demonstrated that even a single agent trained using a relatively small number of examples can robustly solve a broad range of puzzles represented by Cayley graphs of size up to 10145, confirming the generality of the proposed method.
AB - The paper proposes a novel machine learning-based approach to the pathfindingproblem on extremely large graphs. This method leverages diffusion distanceestimation via a neural network and uses beam search for pathfinding. We demonstrate its efficiency by finding solutions for 4x4x4 and 5x5x5 Rubik’s cubes with unprecedentedly short solution lengths, outperforming all available solvers and introducing the first machine learning solver beyond the 3x3x3 case. In particular, it surpasses every single case of the combined best results in the Kaggle Santa 2023 challenge, which involved over 1,000 teams. For the 3x3x3 Rubik’s cube, our approach achieves an optimality rate exceeding 98%, matching the performance of task-specific solvers and significantly outperforming prior solutions such as DeepCubeA (60.3%) and EfficientCube (69.6%). Our solution in its current implementation is approximately 25.6 times faster in solving 3x3x3 Rubik’s cubes while requiring up to 8.5 times less model training time than the most efficient state-of-the-art competitor. Finally, it is demonstrated that even a single agent trained using a relatively small number of examples can robustly solve a broad range of puzzles represented by Cayley graphs of size up to 10145, confirming the generality of the proposed method.
KW - pathfinding
KW - multi-agent
KW - ML
KW - machine learning
KW - Cayley graph
KW - Rubik
KW - cube
KW - MLP
UR - https://www.scopus.com/pages/publications/105048567170
M3 - Conference contribution
T3 - Advances in Neural Information Processing Systems
SP - 191182
EP - 191204
BT - Advances in Neural Information Processing Systems
PB - Neural information processing systems foundation
T2 - 38th Conference on Neural Information Processing Systems
Y2 - 2 December 2025 through 7 December 2025
ER -
