Nature:AI 引导人类直觉,帮助发现数学定理
导语
我们通常认为,数学家的世界充满了直觉和想象力,他们发现模型、提出猜想、证明定理;而计算机只不过是擅长机械的计算。但能够从大量数据中学习的 AI,是否能够像数学家一样,从数据中发现模式?是否可以辅助数学家做出新发现呢?12月1日,DeepMind 团队在 Nature 杂志上发表的一项最新研究中,人们成功让 AI 与人类数学家进行了合作,利用机器学习从大规模数据中探测模式,然后数学家尝试据此提出猜想,精确表述猜想并给出严格证明。他们解决了纯数学领域的两个问题:得到了纽结理论中代数和几何不变量之间的关系,提出了表示论中组合不变性猜想的可能证明方法。这次成功意味着未来机器学习可能会被引入数学家的工作中,AI 和数学家之间将展开更深入的合作。有数学家认为,这就像是伽利略有了望远镜,能够凝视数据宇宙的深处,看到之前从未探测到的东西。以下是这篇论文的翻译。
研究领域:人工智能,机器学习,纽结理论,表示论
Alex Davies, Petar Veličković, Lars Buesing等 | 作者
赵雨亭 | 译者
潘佳栋 | 审校
邓一雪 | 编辑
论文题目:
Advancing mathematics by guiding human intuition with AI
论文链接:https://www.nature.com/articles/s41586-021-04086-x
1. 摘要
1. 摘要
2. 引入
2. 引入
3. 使用AI引导数学直觉
3. 使用AI引导数学直觉
图1. 框架流程图。通过训练机器学习模型来估计特定数据P(Z)分布上的函数,该过程有助于引导数学家对猜想的函数f的直觉。来自学习函数
*注:例如,我们常见的欧拉公式形式为F-E+V=2,即 -V+E+2=F,V、E、F分别表示顶点、边、面的数量。
4. 拓扑学:
纽结理论中代数与几何不变量的关系
4. 拓扑学:
纽结理论中代数与几何不变量的关系
图2. 三个双曲纽结的不变量示例。我们假设几何和代数不变量之间存在先前未发现的关系。
图3. 扭结理论归因。a. 每个输入X(z)的属性值。具有高值的特征是那些学习函数最敏感的特征,并且可能与进一步探索相关。95% 置信区间误差线跨越模型的 10 次重新训练。b. 相关特征的示例可视化——经向平移相对于符号差的实部,由纵向平移着色。
5. 表示论:对称群组合不变性猜想
5. 表示论:对称群组合不变性猜想
图4. 两个示例数据集元素,一个来自S5,一个来自S6。组合不变性猜想指出,一对置换的KL多项式应该可以从它们未标记的Bruhat区间计算出来,但是此前人们并不知道计算的函数。
注:Bruhat 区间是一种图,表示一次只交换两个对象,让集合中的对象逆转顺序的所有不同方式。KL多项式告诉数学家关于这个图在高维空间中存在的不同方式的一些深刻而微妙的性质。只有当 Bruhat 区间有100或1000个顶点时,才会出现有趣的结构。
图 5. 表示论归因。a. 在预测q4时,与数据集中跨区间的平均值相比,显著子图中存在的反射增加百分比的示例热图。b. 与来自数据集的10个相同大小的自举样本相比,模型的10次再训练在显著子图中观察到的每种类型的边缘的百分比。误差线是95%的置信区间,显示的显著性水平是使用双侧双样本t检验确定的。*p < 0.05;****p < 0.0001。c,通过假设、监督学习和归因的迭代过程发现的有趣子结构的区间021435–240513∈S6的说明。受先前工作[31]启发的子图以红色突出显示,超立方体以绿色突出显示,分解成分与SN-1中的区间同构以蓝色突出显示。
6. 结论
6. 结论
参考文献
1. Borwein, J. & Bailey, D. Mathematics by Experiment (CRC, 2008).2. Birch, B. J. & Swinnerton-Dyer, H. P. F. Notes on elliptic curves. II. J. Reine Angew. Math. 1965, 79–108 (1965).3. Carlson, J. et al. The Millennium Prize Problems (American Mathematical Soc., 2006).4. Brenti, F. Kazhdan-Lusztig polynomials: history, problems, and combinatorial invariance. Sémin. Lothar. Combin. 49, B49b (2002).5. Hoche, R. Nicomachi Geraseni Pythagorei Introductions Arithmeticae Libri 2 (In aedibus BG Teubneri, 1866).6. Khovanov, M. Patterns in knot cohomology, I. Exp. Math. 12, 365–374 (2003).7. Appel, K. I. & Haken, W. Every Planar Map Is Four Colorable Vol. 98 (American Mathematical Soc., 1989).8. Scholze, P. Half a year of the Liquid Tensor Experiment: amazing developments Xena https://xenaproject.wordpress.com/2021/06/05/half-a-year-of- the-liquid-tensor- experiment-amazing-developments/ (2021).9. Fajtlowicz, S. in Annals of Discrete Mathematics Vol. 38 113–118 (Elsevier, 1988).10. Larson, C. E. in DIMACS Series in Discrete Mathematics and Theoretical Computer Science Vol. 69 (eds Fajtlowicz, S. et al.) 297–318 (AMS & DIMACS, 2005).11. Raayoni, G. et al. Generating conjectures on fundamental constants with the Ramanujan machine. Nature 590, 67–73 (2021).12. MacKay, D. J. C. Information Theory, Inference and Learning Algorithms (Cambridge Univ. Press, 2003).13. Bishop, C. M. Pattern Recognition and Machine Learning (Springer, 2006).14. LeCun, Y., Bengio, Y. & Hinton, G. Deep learning. Nature 521, 436–444 (2015).15. Raghu, M. & Schmidt, E. A survey of deep learning for scientific discovery. Preprint at https://arxiv.org/abs/2003.11755 (2020).16. Wagner, A. Z. Constructions in combinatorics via neural networks. Preprint at https://arxiv.org/abs/2104.14516 (2021).17. Peifer, D., Stillman, M. & Halpern-Leistner, D. Learning selection strategies in Buchberger’s algorithm. Preprint at https://arxiv.org/abs/2005.01917 (2020).18. Lample, G. & Charton, F. Deep learning for symbolic mathematics. Preprint at https://arxiv.org/abs/1912.01412 (2019).19. He, Y.-H. Machine-learning mathematical structures. Preprint at https://arxiv.org/abs/2101.06317 (2021).20. Carifio, J., Halverson, J., Krioukov, D. & Nelson, B. D. Machine learning in the string landscape. J. High Energy Phys. 2017, 157 (2017).21. Heal, K., Kulkarni, A. & Sertöz, E. C. Deep learning Gauss-Manin connections. Preprint at https://arxiv.org/abs/2007.13786 (2020).22. Hughes, M. C. A neural network approach to predicting and computing knot invariants. Preprint at https://arxiv.org/abs/1610.05744 (2016).23. Levitt, J. S. F., Hajij, M. & Sazdanovic, R. Big data approaches to knot theory: understanding the structure of the Jones polynomial. Preprint athttps://arxiv.org/abs/1912.10086 (2019).24. Jejjala, V., Kar, A. & Parrikar, O. Deep learning the hyperbolic volume of a knot. Phys. Lett. B 799, 135033 (2019).25. Tao, T. There’s more to mathematics than rigour and proofs Blog https://terrytao.wordpress.com/career-advice/theres-more-to-mathematics-than-rigour-and-proofs/ (2016).26. Kashaev, R. M. The hyperbolic volume of knots from the quantum dilogarithm. Lett. Math. Phys. 39, 269–275 (1997).27. Davies, A., Lackenby, M., Juhasz, A. & Tomašev, N. The signature and cusp geometry of hyperbolic knots. Preprint at arxiv.org (in the press).28. Curtis, C. W. & Reiner, I. Representation Theory of Finite Groups and Associative Algebras Vol. 356 (American Mathematical Soc., 1966).29. Brenti, F., Caselli, F. & Marietti, M. Special matchings and Kazhdan–Lusztig polynomials. Adv. Math. 202, 555–601 (2006).30. Verma, D.-N. Structure of certain induced representations of complex semisimple Lie algebras. Bull. Am. Math. Soc. 74 , 160–166 (1968).31. Braden, T. & MacPherson, R. From moment graphs to intersection cohomology. Math. Ann. 321, 533–551 (2001).32. Blundell, C., Buesing, L., Davies, A., Veličković, P. & Williamson, G. Towards combinatorial invariance for Kazhdan-Lusztig polynomials. Preprint at arxiv.org (in the press).33. Silver, D. et al. Mastering the game of Go with deep neural networks and tree search. Nature 529, 484–489 (2016).34. Kanigel, R. The Man Who Knew Infinity: a Life of the Genius Ramanujan (Simon and Schuster, 2016).35. Poincaré, H. The Value of Science: Essential Writings of Henri Poincaré (Modern Library, 1907).36. Hadamard, J. The Mathematician’s Mind (Princeton Univ. Press, 1997).37. Bronstein, M. M., Bruna, J., Cohen, T. & Veličković, P. Geometric deep learning: grids, groups, graphs, geodesics, and gauges. Preprint at https://arxiv.org/abs/2104.13478 (2021).38. Efroymson, M. A. in Mathematical Methods for Digital Computers 191–203 (John Wiley, 1960).39. Xu, K. et al. Show, attend and tell: neural image caption generation with visual attention. In Proc. International Conference on Machine Learning 2048–2057 (PMLR, 2015).40. Sundararajan, M., Taly, A. & Yan, Q. Axiomatic attribution for deep networks. In Proc. International Conference on Machine Learning 3319–3328 (PMLR, 2017).41. Bradbury, J. et al. JAX: composable transformations of Python+NumPy programs (2018); https://github.com/google/jax42. Martín A. B. A. D. I. et al. TensorFlow: large-scale machine learning on heterogeneous systems (2015); https://doi.org/10.5281/zenodo.4724125.43. Paszke, A. et al. in Advances in Neural Information Processing Systems 32 (eds Wallach, H. et al.) 8024–8035 (Curran Associates, 2019).44. Thurston, W. P. Three dimensional manifolds, Kleinian groups and hyperbolic geometry. Bull. Am. Math. Soc 6, 357–381 (1982).45. Culler, M., Dunfield, N. M., Goerner, M. & Weeks, J. R. SnapPy, a computer program for studying the geometry and topology of 3-manifolds (2020); http://snappy.computop.org.46. Burton, B. A. The next 350 million knots. In Proc. 36th International Symposium on Computational Geometry (SoCG 2020) (Schloss Dagstuhl-Leibniz-Zentrum für Informatik, 2020).47. Warrington, G. S. Equivalence classes for the μ-coefficient of Kazhdan–Lusztig polynomials in Sn. Exp. Math. 20, 457–466 (2011). 48. Gilmer, J., Schoenholz, S. S., Riley, P. F., Vinyals, O. & Dahl, G. E. Neural message passing for quantum chemistry. Preprint at https://arxiv.org/abs/1704.01212 (2017).49. Veličković, P., Ying, R., Padovano, M., Hadsell, R. & Blundell, C. Neural execution of graph algorithms. Preprint at https://arxiv.org/abs/1910.10593 (2019).
(参考文献可上下滑动查看)
复杂科学最新论文
集智斑图顶刊论文速递栏目上线以来,持续收录来自Nature、Science等顶刊的最新论文,追踪复杂系统、网络科学、计算社会科学等领域的前沿进展。现在正式推出订阅功能,每周通过微信服务号「集智斑图」推送论文信息。扫描下方二维码即可一键订阅:
推荐阅读
微积分的未来:DNA、非线性、混沌、复杂系统与人工智能 物理网络中的监督学习:从机器学习到学习机器 Nat. Mach. Intell:优化物理学和机器学习之间的协同作用 《张江·复杂科学前沿27讲》完整上线! 成为集智VIP,解锁全站课程/读书会 加入集智,一起复杂!
点击“阅读原文”,追踪复杂科学顶刊论文