Abstract
It is shown for the first time in this paper, that Kleinberg’s (2002) (self-contradictory) axiomatic system for distance-based clustering fails (that is one of the data transforming axioms, consistency axiom, turns out to be identity transformation) in fixed-dimensional Euclidean space due to the consistency axiom limitations and that its replacement with inner-consistency or outer consistency does not help if continuous data transformations are required. Therefore we formulate a new, sound axiomatic framework for cluster analysis in the fixed dimensional Euclidean space, suitable for k-means like algorithms. The system incorporates centric consistency axiom and motion consistency axiom which induce clustering preserving transformations useful e.g. for deriving new labelled sets for testing clustering procedures. It is suitable for continuous data transformations so that labelled data with small perturbations can be derived. Unlike Kleinberg’s consistency, the new axioms do not lead the data outside of Euclidean space nor cause increase in data dimensionality. Our cluster preserving transformations have linear complexity in data transformation and checking. They are in practice less restrictive, less rigid than Kleinberg’s consistency as they do not enforce inter-cluster distance increase and inner cluster distance decrease when performing clustering preserving transformation.
Similar content being viewed by others
Notes
This impossibility does not mean that there is an inner-contradiction when executing the inner-Γ-transform. Rather it means that considering inner-consistency is pointless because inner-Γ-transform is in general impossible except for isometric transformation.
This property holds clearly for k-means, if quality is measured by inverted Q function, but also we can measure cluster quality of k-single-link with the inverted longest link in any cluster and then the property holds.
k-means quality function is known to exhibit local minima at which the k-means algorithm may get stuck at. This claim means that after the centric Γ-transformation a partition will still be a local optimum. If the quality function has a unique local optimum then of course it is a global optimum and after the transform the partition yielding this global optimum will remain the global optimum.
A clustering function clustering into k clusters has the locality property, if whenever a set S for a given k is clustered by it into the partition Γ, and we take a subset \({\Gamma }^{\prime }\subset {\Gamma }\) with \(|{\Gamma }^{\prime }|=k^{\prime }<k\), then clustering of \(\cup _{C\in {\Gamma }^{\prime }}\) into \(k^{\prime }\) clusters will yield exactly \({\Gamma }^{\prime }\).
References
Ackerman M (2012) Towards theoretical foundations of clustering, university of Waterloo. PhD Thesis
Ackerman M, Ben-david S, Loker D (2010) Characterization of linkage-based clustering. In: COLT. pp 270–281
Ackerman M, Ben-David S, Loker D (2010) Towards property-based classification of clustering paradigms. In: Adv neural information proc Sys. 23. pp 10–18, Curran Associates, Inc.
Ackerman M, Ben-David S, Brânzei S, Loker D (2021) Weighted clustering: towards solving the user’s dilemma. Pattern Recogn 120:108152. https://doi.org/10.1016/j.patcog.2021.108152
Awasthi P, Blum A, Sheffet O (2012) Center-based clustering under perturbation stability. Inf Process Lett 112(1-2):49–54
Balcan M, Liang Y (2016) Clustering under perturbation resilience. SIAM J Comput 45 (1):102–155
Ben-David S, Ackerman M (2009) Measures of clustering quality: a working set of axioms for clustering. In: Koller D, Schuurmans D, Bengio Y, Bottou L (eds) Advances in neural information processing systems 21. Curran Associates, Inc., pp 121–128
Campagner A, Ciucci D (2020) A formal learning theory for three-way clustering. In: Davis J, Tabia K (eds) Scalable uncertainty management - 14th international conference, SUM 2020, Bozen-Bolzano, Italy, 23-25 September 2020, proceedings. lecture notes in computer science. Springer, vol 12322, pp 128–140, DOI https://doi.org/10.1007/978-3-030-58449-8_9
Carlsson G, Mémoli F (2008) Persistent clustering and a theorem of j. kleinberg. arXiv:0808.2241
Carlsson G, Mémoli F (2010) Characterization, stability and convergence of hierarchical clustering methods. J Mach Learn Res 11:1425–1470
Chang J C, Amershi S, Kamar E (2017) Revolt: collaborative crowdsourcing for labeling machine learning datasets. In: Mark G, Fussell SR, Lampe C, schraefel MC, Hourcade JP, Appert C, Wigdor D (eds) Proceedings of the 2017 CHI conference on human factors in computing systems, Denver, CO, USA, 06-11 May 2017, pp 2334–2346. ACM, DOI https://doi.org/10.1145/3025453.3026044
Lin C-R, chen M-S (2005) Combining partitional and hierarchical algorithms for robust and efficient data clustering with cohesion self-merging. IEEE Trans Knowl Data Eng 17(2):145– 159
Cohen-Addad V, Kanade V, Mallmann-Trenn F (2018) Clustering redemption beyond the impossibility of kleinberg’s axioms. In: Bengio S, Wallach H, Larochelle H, Grauman K, Cesa-Bianchi N, Garnett R (eds) Advances in neural information processing systems. Curran Associates, Inc., vol 31
Cui X, Yao J, Yao Y (2020) Modeling use-oriented attribute importance with the three-way decision theory. Rough Sets 12179:122–136. https://doi.org/10.1007/978-3-030-52705-1_9
Duda R, Hart P, Stork G (2000) Pattern Classification. Wiley, New York 2nd edn
Gower JC (1990) Clustering axioms. Classification society of North America newsletter. pp 2–3
Hennig C (2015) What are the true clusters? Pattern Recogn Lett 64(15):53–62
Hopcroft J, Kannan R (2012) Computer science theory for the information age. chapter 8.13.2. a satisfiable set of axioms, p 272ff
Iglesias F, Zseby T, Ferreira D (2019) Mdcgen: multidimensional dataset generator for clustering. J Classif 36:599–618. https://doi.org/10.1007/s00357-019-9312-3
Ke Z, Wang D, Yan Q, Ren J, Lau RW (2019) Dual student: breaking the limits of the teacher in semi-supervised learning. In: Proc. IEEE CVF international conference on computer vision. pp 6728–6736
Kleinberg J (2002) An impossibility theorem for clustering. In: Proc. NIPS 2002. pp 446–453. http://books.nips.cc/papers/files/nips15/LT17.pdf
Kłopotek M (2017) On the existence of kernel function for kernel-trick of k-means. In: Kryszkiewicz M, Appice A, §lȩżak D, Rybiński H, Skowron A, Raś Z (eds) Foundations of intelligent systems. ISMIS, Lecture notes in computer science, Springer, Cham. vol 10352
Klopotek MA (2020) On the consistency of k-means++ algorithm. Fundam Inform 172(4):361–377
Kłopotek MA (2022) A clustering preserving transformation for k-means algorithm output
Kłopotek R, Kłopotek M, Wierzchoń S (2020) A feasible k-means kernel trick under non-euclidean feature space. Int J Appl Math Comput Sci 30(4):703–715. https://doi.org/10.34768/amcs-2020-0052
Kłopotek MA, Kłopotek R (2020) In-the-limit clustering axioms. In: To appear in proc. ICAISC2020
Kłopotek MA, Wierzchoń ST, Kłopotek R (2020) k-means cluster shape implications. In: To appear in proc. AIAI
van Laarhoven T, Marchiori E (2014) Axioms for graph clustering quality functions. J Mach Learn Res 15:193–215
Larsen KG, Nelson J, Nguyundefinedn HL, Thorup M (2019) Heavy hitters via cluster-preserving clustering. Commun ACM 62(8):95–100. https://doi.org/10.1145/3339185
Li W, Hannig J, Mukherjee S (2021) Subspace clustering through sub-clusters. J Mach Learn Res 22:1–37
Liu M, Jiang X, Kot AC (2009) A multi-prototype clustering algorithm. Pattern Recognition 42(5):689–698. https://doi.org/10.1016/j.patcog.2008.09.015. http://www.sciencedirect.com/science/article/pii/S0031320308003798
MacQueen J (1967) Some methods for classification and analysis of multivariate observations. In: Proc. Fifth berkeley symp. on math. statist. and prob. vol 1, pp 281–297, Univ. of calif. Press
Moore J, Ackerman M (2016) Foundations of perturbation robust clustering. In: IEEE ICDM. pp 1089–1094, DOI https://doi.org/10.1109/ICDM.2016.0141
Mount DM (2005) Kmlocal: A testbed for k-means clustering algorithms. https://www.cs.umd.edu/mount/Projects/KMeans/kmlocal-doc.pdf
Nie F, Wang CL, Li X (2019) K-multiple-means: A multiple-means clustering method with specified k clusters. In: Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery and data mining. pp 959–967, KDD ’19, association for computing machinery, New York, NY, USA. https://doi.org/10.1145/3292500.3330846
Pollard D (1981) Strong consistency of k–means clustering. Ann Statist 9(1):135–140
Puzicha J, Hofmann T, Buhmann J (2000) A theory of proximity based clustering: structure detection by optimization. Pattern Recogn 33(4):617–634
Qin Y, Ding S, Wang L (2019) Research progress on semi-supervised clustering. Cogn Comput 11:599–612. https://doi.org/10.1007/s12559-019-09664-w
Shekar B (1988) A knowledge-based approach to pattern clustering. Ph.D. thesis Indian Institute of Science
Steinbach M, Karypis G, Kumar V (2000) A comparison of document clustering techniques. In: Proceedings of KDD workshop on text mining, proceedings of the 6th international confer1193 ence on knowledge discovery and data mining, Boston,MA
Strazzeri F, Sánchez-García RJ (2021) Possibility results for graph clustering: a novel consistency axiom. arXiv:1806.06142
Thomann P, Steinwart I, Schmid N (2015) Towards an axiomatic approach to hierarchical clustering of measures
Vidal A, Esteva F, Godo L (2020) Axiomatizing logics of fuzzy preferences using graded modalities. Fuzzy Sets and Systems 401:163–188. https://doi.org/10.1016/j.fss.2020.01.002. https://www.sciencedirect.com/science/article/pii/S0165011419303203,. fuzzy Measures, Integrals and Quantification in Artificial Intelligence Problems – An Homage to Prof. Miguel Delgado
Wang P, Yang X (2021) Three-way clustering method based on stability theory. IEEE Access 9:33944–33953. https://doi.org/10.1109/ACCESS.2021.3057405
Wang X, Kihara D, Luo J, Qi GJ (2021) Enaet: a self-trained framework for semi-supervised and supervised learning with ensemble transformations. IEEE Trans Image Process 30:1639–1647
Wierzchoń S, Kłopotek M (2018) Modern Clustering Algorithms. Studies in Big Data 34. Springer
Wright W (1973) A formalization of cluster analysis. Pattern Rec 5(3):273–282
Zadeh RB, Ben-David S (2009) A uniqueness theorem for clustering. In: Proceedings of the twenty-fifth conference on uncertainty in artificial intelligence. pp 639–646, UAI ’09, AUAI Press, Arlington, Virginia, United States
Zeng G, Wang Y, Pu J, Liu X, Sun X, Zhang J (2016) Communities in preference networks: refined axioms and beyond. In: ICDM. pp 599–608
Zhao Y, Tarus SK, Yang LT, Sun J, Ge Y, Wang J (2020) Privacy-preserving clustering for big data in cyber-physical-social systems: survey and perspectives. Information Sciences 515:132–155. https://doi.org/10.1016/j.ins.2019.10.019. https://www.sciencedirect.com/science/article/pii/S0020025519309764
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Kłopotek, M.A., Kłopotek, R.A. Towards continuous consistency axiom. Appl Intell 53, 5635–5663 (2023). https://doi.org/10.1007/s10489-022-03710-1
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10489-022-03710-1