Article
Computer Science, Artificial Intelligence
Carlo Baldassi
Summary: We introduce an evolutionary algorithm called recombinator-k-means for optimizing the highly nonconvex kmeans problem. Its defining feature is that its crossover step involves all the members of the current generation, stochastically recombining them with a repurposed variant of the k-means++ seeding algorithm. The recombination also uses a reweighting mechanism that realizes a progressively sharper stochastic selection policy and ensures that the population eventually coalesces into a single solution. We compare this scheme with a state-of-the-art alternative, a more standard genetic algorithm with deterministic pairwise-nearest-neighbor crossover and an elitist selection policy, of which we also provide an augmented and efficient implementation. Extensive tests on large and challenging datasets (both synthetic and real word) show that for fixed population sizes recombinator-k-means is generally superior in terms of the optimization objective, at the cost of a more expensive crossover step. When adjusting the population sizes of the two algorithms to match their running times, we find that for short times the (augmented) pairwise-nearest-neighbor method is always superior, while at longer times recombinator-k-means will match it and, on the most difficult examples, take over. We conclude that the reweighted whole-population recombination is more costly but generally better at escaping local minima Moreover, it is algorithmically simpler and more general (it could be applied even to k-medians or k-medoids, for example).
IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION
(2022)
Article
Computer Science, Artificial Intelligence
Yi-Cheng Chen, Yen-Liang Chen, Jyun-Yun Lu
Summary: K-Means algorithm is one of the most famous and popular clustering algorithms in the world, known for its simple structure, easy implementation, high efficiency, and fast convergence speed. This article introduces an improvement to past variants of K-Means used in evolutionary clustering, considering both past and future clustering results, and extending K-Means to multiple cycles, resulting in more consistent, stable, and smooth clustering results.
EXPERT SYSTEMS WITH APPLICATIONS
(2021)
Article
Automation & Control Systems
Uri Stemmer
Summary: This research presents a new algorithm operating in the local model of differential privacy for solving the Euclidean k-means problem, significantly reducing additive error while maintaining multiplicative error. The study shows that the obtained additive error in handling the k-means objective is almost optimal in terms of its dependency on the database size.
JOURNAL OF MACHINE LEARNING RESEARCH
(2021)
Article
Computer Science, Interdisciplinary Applications
Ahmed Fahim
Summary: The k-means method divides N objects into k clusters based on mean values, with linear time complexity and dependence on knowing the number of clusters and initial centers. This research introduces a method able to detect near-optimal values for k and initial centers without prior knowledge, resulting in improved final result quality. The proposed method combines DBSCAN and k-means to converge to global minima and has a time complexity of o(n log n).
JOURNAL OF COMPUTATIONAL SCIENCE
(2021)
Article
Computer Science, Information Systems
Jing Liu, Fuyuan Cao, Jiye Liang
Summary: In this paper, a centroids-guided deep multi-view k-means clustering method is proposed, which incorporates deep representation learning into the multi-view k-means objective. The method produces more k-means-friendly representations by reducing the loss between each representation and its assigned cluster centroid.
INFORMATION SCIENCES
(2022)
Article
Computer Science, Artificial Intelligence
Hongfu Liu, Junxiang Chen, Jennifer Dy, Yun Fu
Summary: K-means is a widely used clustering algorithm known for its simplicity and efficiency. This review paper focuses on generalizing K-means to solve challenging and complex problems. It unifies the available approaches in terms of data representation, distance measure, label assignment, and centroid updating. Concrete applications of modified K-means formulations are reviewed, including iterative subspace projection and clustering, consensus clustering, constrained clustering, domain adaptation, and outlier detection.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Avgoustinos Vouros, Stephen Langdell, Mike Croucher, Eleni Vasilaki
Summary: K-Means is a widely used algorithm for data clustering, but it has limitations such as only finding local minima and being sensitive to initial centroid positions. Various K-Means variations and initialization techniques have been proposed, with more sophisticated techniques reducing the need for complex clustering methods. Deterministic methods generally outperform stochastic methods, but there is a trade-off where simpler stochastic methods run multiple times can result in better clustering.
Article
Computer Science, Artificial Intelligence
Luc Giffon, Valentin Emiya, Hachem Kadri, Liva Ralaivola
Summary: K-means algorithm and Lloyd's algorithm have expanded beyond their original clustering purposes to play pivotal roles in various machine learning and data analysis techniques. QuicK-means is an efficient extension of K-means that reduces computational complexity through sparse matrix products, demonstrating benefits through experimental results.
Article
Computer Science, Artificial Intelligence
Peter Olukanmi, Fulufhelo Nelwamondo, Tshilidzi Marwala
Summary: A key drawback of k-means algorithm is its susceptibility to local minima. The authors propose a technique for comparing initializations directly and selecting the best one based on the maximum minimum inter-center distance. The experiments and mathematical analysis show significant efficiency gains and improved accuracy compared to repeated k-means.
NEURAL COMPUTING & APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Marco Capo, Aritz Perez, Jose A. Antonio
Summary: The K-means algorithm is a popular clustering method, but its performance depends heavily on the initialization phase. Researchers have developed various initialization techniques to address this issue. This article introduces a cost-effective Split-Merge step that can restart the K-means algorithm after reaching a fixed point, reducing error and computing fewer distances.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2022)
Article
Computer Science, Artificial Intelligence
Miaomiao Li, Yi Zhang, Suyuan Liu, Zhe Liu, Xinzhong Zhu
Summary: Multiple kernel clustering (MKC) aims to determine the optimal kernel from several pre-computed basic kernels. A new algorithm called simple multiple kernel k-means with kernel weight regularization (SMKKM-KWR) is proposed to overcome the issue of sparse or over-selected kernel weight coefficients. Experimental results show that SMKKM-KWR achieves effective and efficient clustering performance.
INFORMATION FUSION
(2023)
Article
Computer Science, Information Systems
Simon Harris, Renato Cordeiro De Amorim
Summary: This paper compares the performance of 17 different algorithms on 6,000 synthetic and 28 real-world data sets to investigate the sensitivity of k-means to its initial centroids. The results show that different algorithms may excel in different clustering scenarios, providing valuable insights for those considering k-means for complex clustering tasks.
Article
Computer Science, Interdisciplinary Applications
Rasim M. Alguliyev, Ramiz M. Aliguliyev, Lyudmila Sukhostat
Summary: This article introduces a new parallel batch clustering algorithm based on the k-means algorithm, which reduces computation complexity by splitting the dataset into multiple partitions and proposes a method to determine the optimal batch size. Experimental results show the practical applicability of this method for handling Big Data.
COMPUTERS & INDUSTRIAL ENGINEERING
(2021)
Article
Computer Science, Artificial Intelligence
Chenhui Gao, Wenzhi Chen, Feiping Nie, Weizhong Yu, Feihu Yan
Summary: In this paper, we propose two algorithms, FDKM and IFDKM, for clustering high-dimensional data in a low-dimensional subspace. These algorithms have higher efficiency and lower time complexity compared to traditional methods, and their superior performance is demonstrated in multiple experiments.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Medicine, General & Internal
Kittipol Wisaeng
Summary: This study proposes a new breast cancer detection method based on K-means++ clustering and Cuckoo Search Optimization. By improving the preprocessing and using mathematical morphology, the accuracy and interpretability of the detection are enhanced. Experimental results show that the method achieves an accuracy of over 95% on three datasets, demonstrating its effectiveness.
Article
Computer Science, Artificial Intelligence
Guang-Hai Liu, Zuo-Yong Li, Jing-Yu Yang, David Zhang
Summary: This article introduces a novel image retrieval method that improves retrieval performance by using sublimated deep features. The method incorporates orientation-selective features and color perceptual features, effectively mimicking these mechanisms to provide a more discriminating representation.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Fengguang Peng, Zihan Ding, Ziming Chen, Gang Wang, Tianrui Hui, Si Liu, Hang Shi
Summary: RGB-Thermal (RGB-T) semantic segmentation is an emerging task that aims to improve the robustness of segmentation methods under extreme imaging conditions by using thermal infrared modality. The challenges of foreground-background distinguishment and complementary information mining are addressed by proposing a cross modulation process with two collaborative components. Experimental results show that the proposed method achieves state-of-the-art performances on current RGB-T segmentation benchmarks.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Baihong Han, Xiaoyan Jiang, Zhijun Fang, Hamido Fujita, Yongbin Gao
Summary: This paper proposes a novel automatic prompt generation method called F-SCP, which focuses on generating accurate prompts for low-accuracy classes and similar classes. Experimental results show that our approach outperforms state-of-the-art methods on six multi-domain datasets.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Huikai Liu, Ao Zhang, Wenqian Zhu, Bin Fu, Bingjian Ding, Shengwu Xiong
Summary: Adverse weather conditions present challenges for computer vision tasks, and image de-weathering is an important component of image restoration. This paper proposes a multi-patch skip-forward structure and a Residual Deformable Convolutional module to improve feature extraction and pixel-wise reconstruction.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Oliver M. Crook, Mihai Cucuringu, Tim Hurst, Carola-Bibiane Schonlieb, Matthew Thorpe, Konstantinos C. Zygalakis
Summary: The transportation LP distance (TLP) is a generalization of the Wasserstein WP distance that can be applied directly to color or multi-channelled images, as well as multivariate time-series. TLP interprets signals as functions, while WP interprets signals as measures. Although both distances are powerful tools in modeling data with spatial or temporal perturbations, their computational cost can be prohibitively high for moderate pattern recognition tasks. The linear Wasserstein distance offers a method for projecting signals into a Euclidean space, and in this study, we propose linear versions of the TLP distance (LTLP) that show significant improvement over the linear WP distance in signal processing tasks while being several orders of magnitude faster to compute than the TLP distance.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Haitao Tian, Shiru Qu, Pierre Payeur
Summary: This paper proposes a method of target-dependent classifier, which optimizes the joint hypothesis of domain adaptation into a target-dependent hypothesis that better fits with the target domain clusters through an unsupervised fine-tuning strategy and the concept of meta-learning. Experimental results demonstrate that this method outperforms existing techniques in synthetic-to-real adaptation and cross-city adaptation benchmarks.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Qingsen Yan, Axi Niu, Chaoqun Wang, Wei Dong, Marcin Wozniak, Yanning Zhang
Summary: Deep learning-based methods have achieved remarkable results in the field of super-resolution. However, the limitation of paired training image sets has led researchers to explore self-supervised learning. However, the assumption of inaccurate downscaling kernel functions often leads to degraded results. To address this issue, this paper introduces KGSR, a kernel-guided network that trains both upscaling and downscaling networks to generate high-quality high-resolution images even without knowing the actual downscaling process.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Yifan Chen, Xuelong Li
Summary: Gait recognition is a popular technology for identification due to its ability to capture gait features over long distances without cooperation. However, current methods face challenges as they use a single network to extract both temporal and spatial features. To solve this problem, we propose a two-branch network that focuses on spatial and temporal feature extraction separately. By combining these features, we can effectively learn the spatio-temporal information of gait sequences.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Wei Shi, Wentao Zhang, Wei-shi Zheng, Ruixuan Wang
Summary: This article proposes a simple yet effective visualization framework called PAMI, which does not require detailed model structure and parameters to obtain visualization results. It can be applied to various prediction tasks with different model backbones and input formats.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Xiaobo Hu, Jianbo Su, Jun Zhang
Summary: This paper reviews the latest technologies in pattern recognition, highlighting their instabilities and failures in practical applications. From a control perspective, the significance of disturbance rejection in pattern recognition is discussed, and the existing problems are summarized. Finally, potential solutions related to the application of compensation on features are discussed to emphasize future research directions.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Andres Felipe Posada-Moreno, Nikita Surya, Sebastian Trimpe
Summary: Convolutional neural networks are widely used in critical systems, and explainable artificial intelligence has proposed methods for generating high-level explanations. However, these methods lack the ability to determine the location of concepts. To address this, we propose a novel method for automatic concept extraction and localization based on pixel-wise aggregations, and validate it using synthetic datasets.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Peng Bao, Jianian Li, Rong Yan, Zhongyi Liu
Summary: In this paper, a novel Dynamic Graph Contrastive Learning framework, DyGCL, is proposed to capture the temporal consistency in dynamic graphs and achieve good performance in node representation learning.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Kristian Schultz, Saptarshi Bej, Waldemar Hahn, Markus Wolfien, Prashant Srivastava, Olaf Wolkenhauer
Summary: Research indicates that deep generative models perform poorly compared to linear interpolation-based methods for synthetic data generation on small, imbalanced tabular datasets. To address this, a new approach called ConvGeN, combining convex space learning with deep generative models, has been proposed. ConvGeN improves imbalanced classification on small datasets while remaining competitive with existing linear interpolation methods.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Khondaker Tasrif Noor, Antonio Robles-Kelly
Summary: In this paper, the authors propose H-CapsNet, a capsule network designed for hierarchical image classification. The network effectively captures hierarchical relationships using dedicated capsules for each class hierarchy. A modified hinge loss is utilized to enforce consistency among the involved hierarchies. Additionally, a strategy for dynamically adjusting training parameters is presented to achieve better balance between the class hierarchies. Experimental results demonstrate that H-CapsNet outperforms competing hierarchical classification networks.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Lei Liu, Guorun Li, Yuefeng Du, Xiaoyu Li, Xiuheng Wu, Zhi Qiao, Tianyi Wang
Summary: This study proposes a new agricultural image segmentation model called CS-Net, which uses Simple-Attention Block and Simpleformer to improve accuracy and inference speed, and addresses the issue of performance collapse of Transformers in agricultural image processing.
PATTERN RECOGNITION
(2024)