[Objective/Significance] With the rapid development of information technology, data accumulated in various fields exhibit characteristics such as large scale, diverse types, and complex structures, posing serious challenges to existing unsupervised clustering algorithms. [Method/Process] This article presents a comprehensive survey of various clustering algorithms proposed in recent years. [Results/Conclusions] Based on the different types of data that clustering algorithms can handle, clustering algorithms are categorized into those relying on vector representation and those on relational representation. From a modeling strategy standpoint, clustering algorithms are further stratified into model optimization-based and heuristic-based approaches. Specifically, model optimization-based algorithms delve into the current research status of the k-means algorithm and the graph-cut algorithm, elucidating their distinctions, connections, and the rationale behind the k-means model’s limitation to handling spherical data, while the graph-cut model can accommodate non-convex data. Heuristic-based algorithms are exemplified by density clustering algorithms, which are thoroughly analyzed in this context. In addition, considering the non-convex optimization challenges faced by unsupervised clustering algorithms, this article also analyzes and discusses various optimization methods for unsupervised clustering algorithms. Finally, it summarizes the main characteristics of existing algorithms and optimization methods, and points out the problems existing in the current clustering methods and future research directions.
Xue Jingjing, Chen Huimin, Kong Lingyi, Fan Xinyi, Nie Feiping.
The Foundation of Machine Learning: Current Status and Challenges in Clustering Tasks[J].
Science Focus, 2024, 19(1): 4-17 DOI:10.15978/j.cnki.1673-5668.202401002
其中,$\overline{\mathcal{V}}_{l}$是$\mathcal{G}$中子集$\mathcal{V}_{l}$的补集,$\sigma\left(\mathcal{V}_{l}\right)$为正则项。$\mathbf{L}=\mathbf{D}-\mathbf{W}$是拉普拉斯矩阵,对角阵 D 为度矩阵,其第 i 个对角元为 $d_{i}=\sum_{g} w_{i g}$。
同样,对于锚点图方法,FNC[54]通过引入辅助变量,直接优化基于锚点的图割模型,无需使用 SVD,具有较低的计算复杂度。为了更高效地优化模型,Nie 等人提出了 FDBC[20],只需要使用 CD 更新一个变量。此外,GCSED[59]引入正则化参数,来同时进行谱嵌入和谱旋转。对于大规模平衡图割模型,Labin[60]引了锚点图来扩展SBMC,使之适用于大规模数据;DBSC[61]引了平衡正则化项,以获得锚点图问题的平衡聚类和离散解。表2展示了部分算法。
(3) 讨论
将 k-means 模型写成矩阵形式:
$\min _{\mathbf{Y} \in I n d, \mathbf{M}} \operatorname{Tr}\left(\left(\mathbf{X}-\mathbf{M} \mathbf{Y}^{\mathrm{T}}\right)\left(\mathbf{X}-\mathbf{M} \mathbf{Y}^{\mathrm{T}}\right)^{\mathrm{T}}\right),$
通过对M求导并将导数设为零,得到$\mathbf{M}=\mathbf{X Y}\left(\mathbf{Y}^{\mathrm{T}} \mathbf{Y}\right)^{-1}$,将其带入 k-means 模型中,得到只有一个变量 Y 的模型:
(WangJ, WangS T, DengZ H. Survey on challenges in clustering analysis research[J]. Control and Decision, 2012, 27(3): 321-328.)
[4]
WangX Y, DuY X, YangS, et al. RetCCL: Clustering-guided contrastive learning for whole-slide image retrieval[J]. Medical Image Analysis, 2023, 83: 102645.
[5]
ChenM S, HanJ W, YuP S. Data mining: An overview from a database perspective[J]. IEEE Transactions on Knowledge and Data Engineering, 1996, 8(6): 866-883.
[6]
JiX, VedaldiA, HenriquesJ. Invariant information clustering for unsupervised image classification and segmentation[C]. In Proceedings of the IEEE International Conference on Computer Vision. October 27-November 2, 2019. Seoul, Korea (South). IEEE, 2019, 9865-9874.
[7]
BianchiF M, GrattarolaD, AlippiC. Spectral clustering with graph neural networks for graph pooling[C]. Proceedings of the 37th International Conference on Machine Learning. ACM, 2020, 874-883.
[8]
KarimM R, BeyanO, ZappaA, et al. Deep learning-based clustering approaches for bioinformatics[J]. Briefings in Bioinformatics, 2021, 22(1): 393-415.
[9]
IkotunA M, EzugwuA E, AbualigahL, et al. K-means clustering algorithms: A comprehensive review, variants analysis, and advances in the era of big data[J]. Information Sciences, 2023, 622: 178-210.
[10]
赵兴旺. 大规模复杂数据聚类算法研究[D]. 太原: 山西大学, 2019.
[11]
(ZhaoX W. Research on clustering algorithms for large-scale complex data[D]. Taiyuan: Shanxi University, 2019.)
[12]
MoseleyB, WangJ R. Approximation bounds for hierarchical clustering: Average linkage, bisecting k-means, and local search[J]. Journal of Machine Learning Research, 2023, 24(1): 1-36.
[13]
WuZ, LeahyR. An optimal graph theoretic approach to data clustering: Theory and its application to image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1993, 15(11): 1101-1113.
[14]
KriegelH P, KrögerP, SanderJ, et al. Density-based clustering[J]. WIREs Data Mining and Knowledge Discovery, 2011, 1(3): 231-240.
[15]
TsitsulinA, PalowitchJ, PerozziB, et al. Graph clustering with graph neural networks[J]. Journal of Machine Learning Research, 2023, 24(127): 1-21.
[16]
LiuY, YangX H, ZhouS H, et al. Hard sample aware network for contrastive deep graph clustering[C]. Proceedings of the AAAI Conference on Artificial Intelligence, 2023, 37(7): 8914-8922.
[17]
WangH B, YaoM Z, JiangG Q, et al. Graph-collaborated auto-encoder hashing for multiview binary clustering[J]. IEEE Transactions on Neural Networks and Learning Systems, 2023.
[18]
HartiganJ A, WongM A. Algorithm AS 136: A k-means clustering algorithm[J]. Journal of the Royal Statistical Society, 1979, 28(1): 100-108.
[19]
XueJ J, NieF P, WangR, et al. Iteratively reweighted algorithm for fuzzy c-means[J]. IEEE Transactions on Fuzzy Systems, 2022, 30(10): 4310-4321.
[20]
MurtaghF, ContrerasP. Algorithms for hierarchical clustering: an overview[J]. WIREs Data Mining and Knowledge Discovery, 2012, 2(1): 86-97.
[21]
ChengD F, XuR H, ZhangB, et al. Fast density estimation for density-based clustering methods[J]. Neurocomputing, 2023, 532: 170-182.
[22]
NieF P, XueJ J, WangR, et al. Fast clustering by directly solving bipartite graph clustering problem[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022.
[23]
BanZ H, LiuJ G, CaoL. Superpixel segmentation using gaussian mixture model[J]. IEEE Transactions on Image Processing, 2018, 27(8): 4105-4117.
[24]
LloydS. Least squares quantization in PCM[J]. IEEE Transactions on Information Theory, 1982, 28(2): 129-137.
[25]
WuX D, KumarV, RossQuinlan J, et al. Top 10 algorithms in data mining[J]. Knowledge and Information Systems, 2008, 14(1): 1-37.
[26]
ArthurD, VassilvitskiiS. k-means++: The advantages of careful seeding[C]. In Proceedings of the ACM-SIAM Symposium on Discrete Algorithms, 2006, 1027-1035.
[27]
BachemO, LucicM, HassaniH, et al. Fast and provably good seedings for k-means[C]. In Proceedings of the International Conference on Neural Information Processing System, 2016, 55-63.
[28]
VincentCohen-Addad, SilvioLattanzi, AshkanNorouzi-Fard, et al. Fast and accurate k-means++ via rejection sampling[C]. In Proceedings of the International Conference on Neural Information Processing System, 2020, 16235-16245.
[29]
ElkanC. Using the triangle inequality to accelerate k-means[C]. In Proceedings of the International Conference on Machine Learning, 2003, 147-153.
[30]
XiaS Y, PengD W, MengD Y, et al. A fast adaptive k-means with no bounds[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020.
[31]
DingY F, ZhaoY, ShenX P, et al. Yinyang k-means: A drop-in replacement of the classic k-means with consistent speedup[C]. In International Conference on Machine Learning, PMLR, 2015, 579-587.
[32]
NewlingJ, FleuretF. Fast k-means with accurate bounds[C]. In International Conference on Machine Learning, PMLR, 2016, 936-944.
[33]
NieF P, XueJ J, WuD Y, et al. Coordinate descent method for k k-means[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 44(5): 2371-2385.
[34]
NieF P, LiZ H, WangR, et al. An effective and efficient algorithm for k-means clustering with new formulation[J]. IEEE Transactions on Knowledge and Data Engineering, 2022, 35(4): 3433-3443.
[35]
AhmedM, SerajR, IslamS M S. The k-means algorithm: A comprehensive survey and performance evaluation[J]. Electronics, 2020, 9(8): 1295.
[36]
FräntiP, SieranojaS. How much can k-means be improved by using better initialization and repeats?[J]. Pattern Recognition, 2019, 93: 95-112.
[37]
BachemO, LucicM, HassaniS H, et al. Approximate k-means++ in sublinear time[C]. In Proceedings of the AAAI Conference on Artificial Intelligence, 2016, 1459-1467.
[38]
BachemO, LucicM, KrauseA. Distributed and provably good seedings for k-means in constant rounds[C]. In Proceedings of the International Conference on Machine Learning, 2017, 292-300.
[39]
KantS, Ahmad Ansari I. An improved K means clustering with Atkinson index to classify liver patient dataset[J]. International Journal of System Assurance Engineering and Management, 2016, 7: 222-228.
[40]
BahmaniB, MoseleyB, VattaniA, et al. Scalable k-means++[J]. Proceedings of the VLDB Endowment, 2012, 5(7): 622-633.
[41]
OrchardM T. A fast nearest-neighbor search algorithm. In Acoustics, Speech, and Signal Processing, IEEE International Conference on[C]. IEEE Computer Society, 1991,2297-2298.
[42]
KanungoT, MountD M, NetanyahuN S, et al. An efficient k-means clustering algorithm: Analysis and implementation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2002, 24(7): 881-892.
[43]
PellegD, MooreA. Accelerating exact k-means algorithms with geometric reasoning[C]. In Proceedings of the Fifth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 1999, 277-281.
[44]
NewlingJ, FleuretF. Nested mini-batch k-means[C]. In Proceedings of the International Conference on Neural Information Processing System, 2016, 1352-1360.
[45]
CohenM B, ElderS, MuscoC, et al. Dimensionality reduction for k-means clustering and low rank approximation[C]. In Proceedings of the ACM Symposium on Theory of Computing, 2015, 163-172.
[46]
SinhaK. K-means clustering using random matrix sparsification[C]. In Proceedings of the International Conference on Machine Learning, 2018, 4684-4692.
[47]
HamerlyG. Making k-means even faster[C]. In Proceedings of the SIAM International Conference on Data Mining, 2010, 130-140.
[48]
NewlingJ, FleuretF. K-medoids for k-means seeding[J]. Advances in Neural Information Processing Systems, 2017, 30,.
[49]
NgR T, HanJ W. CLARANS: A method for clustering objects for spatial data mining[J]. IEEE Transactions on Knowledge and Data Engineering, 2002, 14(5): 1003-1016.
[50]
FowlkesC, BelongieS, ChungF, et al. Spectral grouping using the nyström method[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2004, 26(2): 214-225.
[51]
YanD H, HuangL, JordanM I. Fast approximate spectral clustering[C]. In Proceedings of the 15th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2009, 907-916.
[52]
ZhuW, NieF P, LiX L. Fast spectral clustering with efficient large graph construction[C]. In 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, 2017, 2492-2496.
[53]
CaiD, ChenX L. Large scale spectral clustering via landmark-based sparse representation[J]. IEEE Transactions on Cybernetics, 2014, 45(8): 1669-1680.
[54]
ChenX J, NieF P, HuangJ Z X, et al. Scalable normalized cut with improved spectral rotation[C]. In IJCAI, 2017, 1518-1524.
[55]
WangC L, NieF P, WangR, et al. Revisiting fast spectral clustering with anchor graph[C]. In ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, 2020, 3902-3906.
[56]
ChenX J, HongW J, NieF P, et al. Spectral clustering of large-scale data by directly solving normalized cut[C]. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2018, 1206-1215.
[57]
ChenX J, XiaoZ C, NieF P, et al. Finc: An efficient and effective optimization method for normalized cut[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022.
[58]
NieF P, LuJ T, WuD Y, et al. A novel normalized-cut solver with nearest neighbor hierarchical initialization[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023.
[59]
ChenX J, HaungJ Z, NieF P, et al. A self-balanced min-cut algorithm for image clustering[C]. In Proceedings of the IEEE International Conference on Computer Vision, 2017, 2061-2069.
[60]
ChenX J, HongW J, NieF P, et al. Enhanced balanced min cut[J]. International Journal of Computer Vision, 2020, 128:1982-1995.
[61]
WangZ, LiZ Q, WangR, et al. Large graph clustering with simultaneous spectral embedding and discretization[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 43(12): 4426-4440.
[62]
ChenX J, ChenR J, WuQ Y, et al. Labin: Balanced min cut for large-scale data[J]. IEEE Transactions on Neural Networks and Learning Systems, 2019, 31(3): 725-736.
[63]
WangR, ChenH M, LuY H, et al. Discrete and balanced spectral clustering with scalability[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023.
[64]
HagenL, KahngA B. New spectral methods for ratio cut partitioning and clustering[J]. IEEE Transactions on Computer-aided Design of Integrated Circuits and Systems, 1992, 11(9): 1074-1085.
[65]
ShiJ B, MalikJ. Normalized cuts and image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2000, 22(8): 888-905.
[66]
TremblayN, LoukasA. Approximating spectral clustering via sampling: a review[J]. Sampling Techniques for Supervised or Unsupervised Tasks, 2020, 129-183.
[67]
SpielmanD A, TengS H. Spectral sparsification of graphs[J]. SIAM Journal on Computing, 2011, 40(4): 981-1025.
[68]
GittensA, MahoneyM W. Revisiting the nyström method for improved large-scale machine learning[C]. In International Conference on Machine Learning, PMLR, 2013, 567-575.
[69]
BouneffoufD, BirolI. Sampling with minimum sum of squared similarities for nyström-based large scale spectral clustering[C]. In IJCAI, 2015, 2313-2319.
[70]
LiJ, NieF P, LiX L. Directly solving the original ratiocut problem for effective data clustering[C]. In Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, 2018, 2306-2310.
[71]
EsterM, KriegelH P, SanderJ, et al. A density-based algorithm for discovering clusters in large spatial databases with noise[C]. In KDD, 1996, 96: 226-231.
[72]
AnkerstM, BreunigM M, KriegelH P, et al. Optics: Ordering points to identify the clustering structure[J]. ACM Sigmod Record, 1999, 28(2): 49-60.
[73]
McInnesL, HealyJ, AstelsS. hdbscan: Hierarchical density based clustering[J]. The Journal of Open Source Software, 2017, 2(11): 205.
[74]
HinneburgA, GabrielH H. DENCLUE 2.0: Fast clustering based on kernel density estimation[C]. In International Symposium on Intelligent Data Analysis, Springer, 2007, 70-80.
[75]
HouJ, GaoH J, LiX L. DSets-DBSCAN: A parameter-free clustering algorithm[J]. IEEE Transactions on Image Processing, 2016, 25(7): 3182-3193.
[76]
ChenY W, ZhouL D, PeiS W, et al. KNN-BLOCK DBSCAN: Fast clustering for large-scale data[J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2019, 51(6): 3939-3953.
[77]
RodriguezA, LaioA. Clustering by fast search and find of density peaks[J]. Science, 2014, 344(6191): 1492-1496.
[78]
ChenY W, HuX L, FanW T, et al. Fast density peak clustering for large scale data based on KNN[J]. Knowledge-Based Systems, 2020, 187: 104824.
[79]
XuJ, WangG Y, DengW H. DenPEHC: Density peak based efficient hierarchical clustering[J]. Information Sciences, 2016, 373: 200-218.
[80]
GuanJ Y, LiS, HeX X, et al. Fast hierarchical clustering of local density peaks via an association degree transfer method[J]. Neurocomputing, 2021, 455: 401-418.
[81]
QiuT, LiY J. Fast LDP-MST: An efficient density-peak-based clustering method for large-size datasets[J]. IEEE Transactions on Knowledge and Data Engineering, 2022, 35(5): 4767-4780.
[82]
ShiJ L, LuoZ G. A novel clustering approach based on the manifold structure of gene expression data[C]. In 2010 4th International Conference on Bioinformatics and Biomedical Engineering, IEEE, 2010, 1-4.
[83]
SongY, GuY, ZhangR, et al. Brepartition: Optimized high-dimensional kNN search with bregman distances[J]. IEEE Transactions on Knowledge and Data Engineering, 2020, 34(3): 1053-1065.
[84]
PavanM, PelilloM. Dominant sets and pairwise clustering[J]. IEEE transactions on pattern analysis and machine intelligence, 2006, 29(1): 167-172.
[85]
ChenJ W, XieS Y, JiangH Y, et al. A novel k-means framework via constrained relaxation and spectral rotation[J]. IEEE Transactions on Neural Networks and Learning Systems, 2023.
[86]
ChenQ, YuW Z, ZhaoX W, et al. Rooted mahalanobis distance based gustafson-kessel fuzzy c-means[J]. Information Sciences, 2023, 644: 118878.
[87]
NieF P, XueJ J, YuW Z, et al. Fast clustering with anchor guidance[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023.
[88]
ChakrabortyS, MaliK. Fuzzy modified cuckoo search for biomedical image segmentation[J]. Knowledge and Information Systems, 2022, 64(4): 1121-1160.
[89]
ZhangZ C, YangJ L. A discrete cuckoo search algorithm for traveling salesman problem and its application in cutting path optimization[J]. Computers and Industrial Engineering, 2022, 169: 108157.
[90]
GuoY Q, GuoJ F, SunB Z, et al. A new decomposition ensemble model for stock price forecasting based on system clustering and particle swarm optimization[J]. Applied Soft Computing, 2022, 130: 109726.
(HeX S, ChenH Y. Robot path planning based on obstacle avoidance optimization and improved ant colony algorithm[J/OL]. Journal of Xi’an Polytechnic University, 2024, 38(2): 1-9.)
JiangD R, XuS C. Large data clustering processing method based on improved PSO means clustering algorithm[[J/OL]]. Journal of Jilin University (Information Science Edition, 1-8. https://doi.org/10.19292/j.cnki.jdxxp.20231128.002)
(GuJ X, HeX S, LiuQ. Parameter optimization of support vector machine based on improved cuckoo search algorithm[J]. Journal of Xi’an Polytechnic University, 2022, 36(2): 110-118.)
[102]
SelimS Z, AlsultanK. A simulated annealing algorithm for the clustering problem[J]. Pattern Recognition, 1991, 24(10): 1003-1008.
[103]
MuritibaA E F, GomesM J N, de SouzaM F, et al. Path-relinking with tabu search for the capacitated centered clustering problem[J]. Expert Systems with Applications, 2022, 116766.
[104]
DiMartino F, SessaS. A novel quantum inspired genetic algorithm to initialize cluster centers in fuzzy c-means[J]. Expert Systems with Applications, 2022, 191: 116340.
[105]
CuraT. A particle swarm optimization approach to clustering[J]. Expert Systems with Applications, 2012, 39(1): 1582-1588.
[106]
HuH Z, LiuJ X, ZhangX P, et al. An effective and adaptable k-means algorithm for cluster analysis[J]. Pattern Recognition, 2023, 139: 109404.