模块1: 追根溯源:互信息、有效信息、动力学约简 【简介】因果涌现理论为我们解释涌现提供了新的视角,人们发现,对于具有涌现性质的系统,其宏观尺度会比微观尺度具有更明确的因果关系,而有效信息是衡量着一因果的手段,有效信息在本质上是一种互信息,这一部分我们将追根溯源,去询问这一问题:从信息论的视角来看,有效信息增加的这一过程中到底发生了什么?在这一模块我们将涉及到互信息的分解,认识协同信息,冗余信息,也将涉及到整合信息等话题。 【参考文献】1.Decomposition of mutual information[1] Williams PL, Beer RD. Nonnegative decomposition of multivariate information. arXiv preprint arXiv:10042515. 2010;.[2] Timme N, Alford W, Flecker B, Beggs JM. Synergy, redundancy, and multivariate information measures: An experimentalist's perspective. Journal of Computational Neuroscience. 2014;36(2):119–140. pmid:23820856[3] McGill WJ. Multivariate information transmission. Psychometrika. 1954;19(2):97–116.(early)[4] Mediano PA, Rosas F, Carhart-Harris RL, Seth AK, Barrett AB. Beyond integrated information: A taxonomy of information dynamics phenomena. arXiv preprint arXiv:190902297. 2019;.[5] Rosas F, Mediano P, Rassouli B, Barrett A. An operational information decomposition via synergistic disclosure. arXiv preprint arXiv:200110387. 2020;.[6] Rassouli B, Rosas FE, Gündüz D. Data Disclosure under Perfect Sample Privacy. IEEE Transactions on Information Forensics and Security. 2019;.(Algorithm)[7] Rosas FE, Mediano PAM, Gastpar M, Jensen HJ. Quantifying high-order interdependencies via multivariate extensions of the mutual information. Physical Review E. 2019(O-information \Omega)[8] Rosas F, Mediano PAM, Ugarte M, Jensen HJ. An information-theoretic approach to self-organisation: Emergence of complex interdependencies in coupled dynamical systems. Entropy. 2018;20(10). pmid:33265882[9] Lizier JT, Bertschinger N, Jost J, Wibral M. Information decomposition of target effects from multi-source interactions: Perspectives on previous, current and future work. Entropy. 2018;20(4). pmid:33265398[10] Modes of information flows, arXiv:1808.06723v1[11] An operational information decomposition via synergistic disclosure, arXiv:2001.10387v2[12] Information Flows? A Critique of Transfer Entropies,PRL 116, 238701 (2016)2.Integrated information theory[1] Tononi, G., Boly, M., Massimini, M. et al. Integrated information theory: from consciousness to its physical substrate. Nat Rev Neurosci 17, 450–461 (2016). https://doi.org/10.1038/nrn.2016.44 Nature综述[2] Giulio Tononi and Olaf Sporns: Measuring information integration, BMC Neuroscience 2003, 4:31[3] Measuring Integrated Information: Comparison of Candidate Measures in Theory and Simulation, https://www.mdpi.com/1099-4300/21/1/17/htm[4] Sara Imari Walker, Hyunju Kim and Paul C. W. Davies,The informational architecture of the cell, Phil. Trans. R. Soc. A 374: 20150057.[5] Krzysztof Chalupka, Pietro Perona, Frederick Eberhardt: Multi-Level Cause-Effect Systems, Proceedings of the 19th International Conference on Artificial Intelligence and Statistics (AISTATS) 2016[6] Chang AY, Biehl M, Yu Y, Kanai R. Information Closure Theory of Consciousness. arXiv preprint arXiv:190913045. 20193.Information Theory and Renormalization[1] Amit Gordon, Aditya Banerjee, Maciej Koch-Janusz, and Zohar Ringel: Relevance in the Renormalization Group and in Information Theory, arXiv:2012.01447v1[2] Giorgio Nicoletti and Daniel Maria Busiello: Mutual information disentangles interactions from changing environments, arXiv:2107.08985v24.Computational Mechanics[1]Cosma Rohilla Shalizi James P. Crutchfield: Computational Mechanics: Pattern and Prediction, Structure and Simplicity,SFI working paper, 1999, https://sfi-edu.s3.amazonaws.com/sfi-edu/production/uploads/sfi-com/dev/uploads/filer/28/93/289376d7-3479-4dba-8acc-90605832399d/99-07-044.pdf5.Synergy or Reduction of Dynamics[1]Bastian Pietras, Andreas Daffertshofer: Network dynamics of coupled oscillators and phase reduction techniques, Physics Reports 819 (2019) 1–105 模块2: 因果与涌现 【简介】近年来,因果涌现理论异军突起,它将因果作为一种定量化的工具来刻画含糊不清的涌现概念。本模块将关注因果涌现和涌现定量化度量这两个方向的进展。在因果涌现方面,如何拓展有效信息工具,让它适用于连续动力学系统是一个决定着因果涌现理论是否能够长足发展的问题;另一方面,从因果涌现的角度可以尝试探索复杂系统中的自上而下的因果,这为我们破解生命乃至意识的谜题提供了理论基础。 【参考文献】[1] Ross Griebenow, Brennan Klein, and Erik Hoel: Finding the right scale of a network: Efficient identification of causal emergence in preferential attachment networks through spectral clustering, arXiv:1908.07565v2, arXiv:2003.13075v1[2] Thomas F. Varley: Causal Emergence in Discrete & Continuous Dynamical Systems[3] Hiroshi Ashikaga,Francisco Prieto-Castrillo, Mari Kawakatsu, and Nima Dehghani: Causal Unit of Rotors in a Cardiac System, arXiv:1712.02203v1[4] Rosas F E, Mediano P A M, Jensen H J, et al. Reconciling emergences: An information-theoretic approach to identify causal emergence in multivariate data[J]. PLoS computational biology, 2020, 16(12): e1008289.https://github.com/pmediano/ReconcilingEmergences[5] Greater than the parts: A review of the information decomposition approach to causal emergence,https://arxiv.org/abs/2111.06518 综述[6] Seth AK. Measuring autonomy and emergence via Granger causality. Artificial Life. 2010;16(2):179–196. pmid:20067405[7] Toy models of topdown causality, https://arxiv.org/abs/1909.12739 [8] Top Down Causality(小木球的读书会) 模块3: 多尺度因果学习 【简介】我们用各种各样的模型来描述世界,如果说基于i.i.d.数据的统计模型是发现表面现象和关联的一端,那么可以处理广泛的o.o.d.、反事实等问题的以偏微分方程为代表的物理模型就是揭示内在规律和因果的另一端。智能现象也是如此。人类有基于重复训练(或演化)的端到端黑盒化的输入-输出反馈型“系统一”智能,也有基于意识和注意力的可对符号和概念进行结构化、分析、推理的“系统二”智能。大量迹象表明,对生命与智能等复杂系统来说,在不同的尺度上具有不同的因果规律,这些规律可以由不同层次的因果模型来表征,在历史上由一代代的科学家来完成,形成了物理学、生物学、心理学等等学科。机器学习为我们进一步发现和表征这些因果模型带来了新的可能性。尽管这个领域的工作还比较少,但我们希望通过一些大佬级科学家的指向性的论文的解读,以及较为深入的讨论,加深对多尺度因果学习的理解,启发这个领域的研究思路,促成科研工作的进展和合作。 【参考文献】[1] Chalupka K, Eberhardt F, Perona P. Causal feature learning: an overview. Behaviormetrika, 2017, 44(1): 137-164.[2] Schölkopf B, Locatello F, Bauer S, et al. Toward causal representation learning. Proceedings of the IEEE, 2021, 109(5): 612-634[3] Benedikt Höltgen: Encoding Causal Macrovariables, arXiv:2111.14724v1[4] Gherardo Varando, Miguel-Angel Fern ´ andez-Torres ´, Gustau Camps-Valls: Learning Granger Causal Feature Representations[5] Mingyu Ding et al. "Dynamic Visual Reasoning by Learning Differentiable Physics Models from Video and Language" neural information processing systems (2021): n. pag. 模块4: 机器学习多尺度自动建模 【简介】涌现与因果之间,具有十分深刻的关联。而这种关联,则可以分别从「多尺度」和「建模」两个角度阐释。在安德森著名的《多者异也》一文中,他点出了不同尺度之间对称性的差异——系统在更大尺度上行为的对称性,往往可以与其微观规律不同。换句话说,不同的尺度,可以有不同的有效理论。而「因果涌现」理论也发现,粗粒化之后的系统,即便忽略了很多信息,也有可能拥有更高的「因果信息」。此二者,在尺度的视角下将涌现与因果关联了起来。与此同时,因果本身就是一个非常有力的建模工具。即便没有具体的数学公式,仅仅靠因果图,我们也可以得到非常丰富的推论。有意思的是,涌现行为常常表现出特别的因果关系——微观上可以有一套因果,宏观上可以有另一套因果。就如我们同时拥有物理、化学、生物一样,我们可以对同一套系统提出多种建模。由此,涌现、因果、建模之间的暗线得以浮出水面。而这一模块的目的,就是希望能通过阅读三个领域的文献,找到它们之间切实的关联,并帮助我们更好地理解这些概念。 【参考文献】[1] J. Zhang: Neural Information Squeezer for Causal Emergence, https://arxiv.org/pdf/2201.10154.pdf[2] Vlachas, P.R., Arampatzis, G., Uhler, C. et al. Multiscale simulations of complex systems by learning their effective dynamics. Nat Mach Intell (2022). https://doi.org/10.1038/s42256-022-00464-w[3] R. Iten, T. Metger, H. Wilming, L. del Rio, R. Renner, Discovering physical concepts with neural networks, Physical Review Letters (2020) 010508[4] M. Kramer, Nonlinear principal component analysis using autoassociative neural networks, AIChE journal 37 (2) (2013) 233–243.[5] Cai, L., & Ji, S.: A Multi-Scale Approach for Graph Link Prediction. Proceedings of the AAAI Conference on Artificial Intelligence, 34(04), 3308-3315, 2020[6] Chen, W.et al.: Multi-Range Attentive Bicomponent Graph Convolutional Network for Traffic Forecasting. Proceedings of the AAAI Conference on Artificial Intelligence, 34(04), 3529-3536, 2020[7] Chen Z. et al.: Multi-Scale Spatial Temporal Graph Convolutional Network for Skeleton-Based Action Recognition, Proceedings of the AAAI Conference on Artificial Intelligence, 35(2), 1113-1122, 2021[8] Rao H. et al.: Multi-Level Graph Encoding with Structural-Collaborative Relation Learning for Skeleton-Based Person Re-Identification, Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence, 973-980, 2021[9] Wang S. et al.: MT-STNets: Multi-Task Spatial-Temporal Networks for Multi-Scale Traffic Prediction, 504 – 512, 2021 模块五:量子因果 【简介】量子因果是一项正在发展中的新兴研究领域,并没有严格的定义。一方面,从因果理论发展本身来看,研究者通过对Judea Pearl因果理论的量子化,讨论量子公因(quantum common cause)和量子因果模型(quantum causal model);另一方面,现代理论物理学中的基本统一理论倾向下,对量子时空或者量子引力的进一步探求,也引发了对量子因果的思考。从某种意义上来说,两种思路,某种意义上也实现了一定程度的殊途同归,因为两种思考下的理论体系有趋于一致的倾向,带来了诸如循环量子因果模型(cyclic quantum causal model)等的新方向。 【参考文献】[1] Allen, John-Mark A., Jonathan Barrett, Dominic C. Horsman, Ciarán M. Lee, and Robert W. Spekkens. "Quantum common causes and quantum causal models." Physical Review X 7, no. 3 (2017): 031021.[2] Barrett, Jonathan, Robin Lorenz, and Ognyan Oreshkov. "Quantum causal models." arXiv preprint arXiv:1906.10726 (2019).[3] Costa, Fabio, and Sally Shrapnel. "Quantum causal modelling." New Journal of Physics 18, no. 6 (2016): 063032.[4] Brukner, Časlav. "Quantum causality." Nature Physics 10, no. 4 (2014): 259-263.[5] Baumeler, Ämin. "Causal loops: logically consistent correlations, time travel, and computation", PhD thesis, Università della Svizzera italiana, (2017).[6] Barrett, Jonathan, Robin Lorenz, and Ognyan Oreshkov. "Cyclic quantum causal models." Nature communications 12, no. 1 (2021): 1-15.[7] Jordan Cotler, Xizhi Han, Xiao-Liang Qi, Zhao Yang. Quantum Causal Influence. https://arxiv.org/pdf/1811.05485.pdf[8] Časlav Brukner. Quantum Causality. https://www.nature.com/articles/nphys2930 [9] Sean Carroll. The Arrow of Time in Causal Networks. https://www.youtube.com/watch?v=6slug9rjaIQ[10] John F. Donoghue, Gabriel Menezes. Quantum causality and the arrows of time and thermodynamics. https://arxiv.org/pdf/2003.09047.pdf