Search

Clustering analysis aims to group unlabeled data in an unsupervised learning manner. However, most existing methods are tailored for single-task data and do not work for a sequence of task...

A new approach to the analysis of heterogeneous categorical sequences is proposed. The first-order Markov model is employed in a finite mixture setting with initial state and transition probabilities being expressed as functions of time. The expectation–maximization algorithm approach to parameter estimation is implemented in the presence of positive equivalence constraints that determine which observations must be placed in the same class in the solution. The proposed model is applied to a dataset from the British Household Panel Survey to evaluate the association between the education background and life outcomes of study participants. The analysis of the survey data reveals many interesting relationships between the level of education and major life events.

Semi.-supervised clustering algorithms aim to enhance the performance of clustering using the pairwise constraints. However, selecting these constraints randomly or improperly can minimize the performance of clustering in certain situations and with different applications. In this paper, we select the most informative constraints to improve semi-supervised clustering algorithms. We present an active selection of constraints, including active must.-link (AML) and active cannot.-link (ACL) constraints. Based on Radial-Bases Function, we compute lower-bound and upper-bound between data points to select the constraints that improve the performance. We test the proposed algorithm with the base-line methods and show that our proposed active pairwise constraints outperform other algorithms.

With the development of new internet application technologies, cyberspace security is becoming more important. How to effectively identify network attacks is the core issue of cyberspace security. Deep learning is used in intrusion detection, which can find hidden attacks in intrusion data and then improve the accuracy of detection. Semi-supervised learning uses a small number of labeled data and a large number of unlabeled data to train. It reduces the requirements of the sample. In this paper, an intrusion detection algorithm based on semi-supervised learning and deep learning is proposed to solve the problem of low accuracy in intrusion detection systems. The algorithm uses sparse self-encoder and softmax classifier in deep learning to classify the data and improves the classification performance. Experimental verification is carried out using KDD CUP99 dataset. The experimental results verify the effectiveness of the algorithm.

The paper discusses a new approach for incorporating hard constraints into the K-means algorithm for semi-supervised clustering. An analytic modification of the objective function of K-means is proposed that has not been previously considered in the literature.

Although semi-supervised clustering ensemble methods have achieved satisfactory performance, they fail to effectively utilize the constrained knowledge such as cannot-link and must-link when generating diverse ensemble members. In addition, they ignore negative effects brought about by redundancies and noisy data. To address the above shortcomings, in this paper we propose an approach to combine multiple semi-supervised clustering solutions via adaptively regularizing the weights of clustering ensemble members, which is referred to as ARSCE. First, we generate a series of feature subspaces by randomly selecting feature without replacement to avoid the scenario where there are two identical feature subspaces. Second, we conduct feature transformation on the above obtained feature subspaces while considering the pairwise constraints to find new clustering-friendly spaces, where clustering methods are exploited to generate various clustering solutions. Finally, we design a novel fusion strategy to integrate multiple clustering solutions into a unified clustering partition, where weights are designated for each clustering ensemble member. Extensive experiments are conducted on multiple real-world benchmarks, and experimental results demonstrate the effectiveness and superiority of our proposed method ARSCE over other counterparts.

In this paper, we propose self-semi-supervised clustering, a new clustering method for large scale data with a massive null group. Self-semi-supervised clustering is a two-stage procedure: preselect a part of “null” group from the data in the first stage and apply semi-supervised clustering to the rest of the data in the second stage, allowing them to be assigned to the null group. We evaluate the performance of the proposed method using a simulation study and demonstrate the method in the analysis of time course gene expression data from a longitudinal study of Influenza A virus infection.

Semi-supervised learning is drawing increasing attention in the era of big data, as the gap between the abundance of cheap, automatically collected unlabeled data and the scarcity of labeled data that are laborious and expensive to obtain is dramatically increasing. In this paper, we first introduce a unified view of density-based clustering algorithms. We then build upon this view and bridge the areas of semi-supervised clustering and classification under a common umbrella of density-based techniques. We show that there are close relations between density-based clustering algorithms and the graph-based approach for transductive classification. These relations are then used as a basis for a new framework for semi-supervised classification based on building-blocks from density-based clustering. This framework is not only efficient and effective, but it is also statistically sound. In addition, we generalize the core algorithm in our framework, HDBSCAN*, so that it can also perform semi-supervised clustering by directly taking advantage of any fraction of labeled data that may be available. Experimental results on a large collection of datasets show the advantages of the proposed approach both for semi-supervised classification as well as for semi-supervised clustering.

Semi-supervised clustering is a new learning method which combines semi-supervised learning (SSL) and cluster analysis. It is widely valued and applied to machine learning. Traditional unsupervised clustering algorithm based on data partition does not need any property; however, there are a small amount of independent class labels or pair constraint information data samples in practice; in order to obtain better clustering results, scholars have proposed a semi-supervised clustering. Compared with traditional clustering methods, it can effectively improve clustering performance through a small number of supervised information, and it has been used widely in machine learning. Firstly, this paper introduces the research status and classification of semi-supervised learning and compares the four classification methods as follows: decentralized model, support vector machine, graph, and collaborative training. Secondly, the semi-supervised clustering is described in detail, the current status of semi-supervised clustering is analyzed, and the Cop-kmeans algorithm, Lcop-kmeans algorithm, Seeded-kmeans algorithm, SC-kmeans algorithm, and other algorithms are introduced. The introduction of several semi-supervised clustering methods in this paper can show the advantages of semi-supervised clustering over traditional clustering, and the related literature in recent years is summarized. This paper summarized the latest development of semi-supervised learning and semi-supervised clustering and discussed the application of semi-supervised clustering and the future research direction.

Cluster analysis methods seek to partition a data set into homogeneous subgroups. It is useful in a wide variety of applications, including document processing and modern genetics. Conventional clustering methods are unsupervised, meaning that there is no outcome variable nor is anything known about the relationship between the observations in the data set. In many situations, however, information about the clusters is available in addition to the values of the features [2]. For example, the cluster labels of some observations may be known, or certain observations may be known to belong to the same cluster. In other cases, one may wish to identify clusters that are associated with a particular outcome variable. This review describes several clustering algorithms (known as �semi-supervised clustering� methods) that can be applied in these situations [3]. The majority of these methods are modifications of the popular k-means clustering method, and several of them will be described in detail. A brief description of some other semi-supervised clustering algorithms is also provided. Cluster formation has three types as supervised clustering, unsupervised clustering and semi supervised. This paper reviews traditional and state-of-the-art methods of clustering [1]. Clustering algorithms are based on active learning, with ensemble clustering-means algorithm, data streams with flock, fuzzy clustering for shape annotations, Incremental semi supervised clustering, Weakly supervised clustering, with minimum labeled data, self-organizing based on neural networks. Incremental semi-supervised clustering ensemble framework (ISSCE) which makes utilization of the advantage of the arbitrary subspace method, the limitation spread approach, the proposed incremental ensemble member choice process, and the normalized cut algorithm to perform high dimensional information clustering [4]. Semi-supervised clustering employs limited supervision in the form of labeled instances or pairwise instance constraints to aid unsupervised clustering and often significantly improves the clustering performance. Despite the vast amount of expert knowledge spent on this problem, most existing work is not designed for handling high-dimensional sparse data.

In this paper, we propose confidence-weighted safe semi-supervised clustering where prior knowledge is given in the form of class labels. In some applications, some samples may be wrongly labeled by the users. Therefore, our basic idea is that different samples should have different impacts or confidences on the clustering performance. In our algorithm, we firstly use unsupervised clustering to perform the dataset partition and compute the normalized confusion matrix N c . N c is used to estimate the safe confidence of each labeled sample based on the assumption that a correctly clustered sample should have a high confidence. Then we construct a local graph to model the relationship between the labeled and its nearest unlabeled samples through the clustering results. Finally, a confidence-weighted fidelity term and a graph-based regularization term are incorporated into the objective function of unsupervised clustering. In this case, on the one hand, the outputs of the labeled samples with high confidences are restricted to be the given prior labels. On the other hand, the outputs of the labeled ones with low confidences are forced to approach those of the local homogeneous unlabeled neighbors modeled by the local graph. Hence, the labeled samples are expected to be safely exploited which is the goal of safe semi-supervised clustering. To verify the effectiveness of our algorithm, we carry out some experiments over several datasets by comparison to the unsupervised and semi-supervised clustering methods and achieve the promising results.

With the rapid increase in the number of collected data samples, semi-supervised clustering (SSC) has become a useful mining tool to find an intrinsic data structure with the help of prior knowledge. The common used prior knowledge includes pair-wise constraints and cluster labels. In the past decades, many relevant methods are proposed to improve clustering performance of SSC by mining prior knowledge. In general, the prior knowledge is assumed to be beneficial to yielding desirable results. However, one can gather inappropriate prior knowledge in some scenarios, such as wrong cluster labels. In this case, prior knowledge can result in degenerating clustering performance. Therefore, how to raise safe semi-supervised clustering (S3C) should be investigated. A main goal of S3C is that the corresponding result is never inferior to that of the corresponding unsupervised clustering part. To achieve the goal, we propose safe semi-supervised Fuzzy c -Means clustering (S3FCM) which is extended from traditional semi-supervised FCM (SSFCM). In our algorithm, wrongly labeled samples are carefully explored by constraining the corresponding predictions to be those yielded by unsupervised clustering. Meanwhile, the predictions of the other labeled samples should approach to the given labels. Therefore the labeled samples are expected to be safely explored through a balance between unsupervised clustering and SSC. From the reported clustering results on different datasets, we can find that S3FCM can yield comparable, if not the best, performance among different unsupervised clustering and SSC methods even if the wrong ratio achieves 20%.

Constraint-based clustering leverages user-provided constraints to produce a clustering that matches the user's expectation. In active constraint-based clustering, the algorithm selects the most informative constraints to query in order to produce good clusterings with as few constraints as possible. A major challenge in constraint-based clustering is handling noise: the majority of existing approaches assume that the provided constraints are correct, while that might not be the case. In this paper, we propose a method to identify and correct noisy constraints in active constraint-based clustering. Our approach reasons probabilistically about the correctness of the user’s answers and asks additional constraints to corroborate or correct the suspicious answers. We demonstrate the method’s effectiveness by incorporating it into COBRAS, a state-of-the-art method for active constraint-based clustering. Compared to COBRAS and other active-constraint-based clustering algorithms, the resulting system produces better clusterings in the presence of noise. ispartof: pages:121-136 ispartof: ECML PKDD 2020: Machine Learning and Knowledge Discovery in Databases vol:12458 pages:121-136 ispartof: European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases location:Ghent, Belgium date:14 Sep - 18 Sep 2020 status: published

Label information in the corresponding semi-supervised fuzzy clustering cannot be used efficiently due to feature redundancy. To address the problem, we propose an adaptive approach of feature selection applied to semi-supervised fuzzy clustering. There are three phases in our approach: 1) feature-score by fisher-score; 2) Min Mean Square Error of Feature Select criterion; 3) the number of features is selected by Min Mean Square Error of Feature Select criterion. We apply our approach to three semi-supervised fuzzy clustering methods. Experiments show that the adaptive approach of feature selection applied to semi-supervised fuzzy clustering can improve the clustering performance.

Many unsupervised methods are widely used for parcellating the brain. However, unsupervised methods aren’t able to integrate prior information, obtained from such as exiting functional neuroanatomy studies, to parcellate the brain, whereas the prior information guided semi-supervised method can generate more reliable brain parcellation. In this study, we propose a novel semi-supervised clustering method for parcellating the brain into spatially and functionally consistent parcels based on resting state functional magnetic resonance imaging (fMRI) data. Particularly, the prior supervised and spatial information is integrated into spectral clustering to achieve reliable brain parcellation. The proposed method has been validated in the hippocampus parcellation based on resting state fMRI data of 20 healthy adult subjects. The experimental results have demonstrated that the proposed method could successfully parcellate the hippocampus into head, body and tail parcels. The distinctive functional connectivity patterns of these parcels have further demonstrated the validity of the parcellation results. The effects of aging on the three hippocampus parcels’ functional connectivity were also explored across the healthy adult subjects. Compared with state-of-the-art methods, the proposed method had better performance on functional homogeneity. Furthermore, the proposed method had good test–retest reproducibility validated by parcellating the hippocampus based on three repeated resting state fMRI scans from 24 healthy adult subjects.

Motivation The rapid development of single-cell RNA sequencing (scRNA-seq) technologies allows us to explore tissue heterogeneity at the cellular level. The identification of cell types plays an essential role in the analysis of scRNA-seq data, which, in turn, influences the discovery of regulatory genes that induce heterogeneity. As the scale of sequencing data increases, the classical method of combining clustering and differential expression analysis to annotate cells becomes more costly in terms of both labor and resources. Existing scRNA-seq supervised classification method can alleviate this issue through learning a classifier trained on the labeled reference data and then making a prediction based on the unlabeled target data. However, such label transference strategy carries with risks, such as susceptibility to batch effect and further compromise of inherent discrimination of target data. Results In this article, inspired by unsupervised domain adaptation, we propose a flexible single cell semi-supervised clustering and annotation framework, scSemiCluster, which integrates the reference data and target data for training. We utilize structure similarity regularization on the reference domain to restrict the clustering solutions of the target domain. We also incorporates pairwise constraints in the feature learning process such that cells belonging to the same cluster are close to each other, and cells belonging to different clusters are far from each other in the latent space. Notably, without explicit domain alignment and batch effect correction, scSemiCluster outperforms other state-of-the-art, single-cell supervised classification and semi-supervised clustering annotation algorithms in both simulation and real data. To the best of our knowledge, we are the first to use both deep discriminative clustering and deep generative clustering techniques in the single-cell field. Availabilityand implementation An implementation of scSemiCluster is available from https://github.com/xuebaliang/scSemiCluster. Supplementary information Supplementary data are available at Bioinformatics online.

In this paper, we present a technique of definitely addressing the pairwise constraints in the semi-supervised clustering. Our method contributes to formulating the cannot-link relations and propagating them over the affinity graph flexibly. The pairwise constrained instances are provably guaranteed to be in the same or different connected components of the graph. Combined with the Laplacian rank constraint, the proposed model learns a Pairwise Constrained structured Optimal Graph (PCOG), from which the specified c clusters supporting the known pairwise constraints are directly obtained. An efficient algorithm invoked by the label propagation is designed to solve the formulation. Additionally, we also provide a compact criterion to acquire the key pairwise constraints for prompting the semi-supervised graph clustering. Substantial experimental results show that the proposed method achieves the significant improvements by using a few prior pairwise constraints.

Social networks permit speedy disperse of innovations and notions despite the fact that rumor or negative information can also spread extensively. Bearing in mind, about its significance and impact, in addition to its complexity, much of the attention of researchers has been pulled by rumor. Theme of online social networks like twitter circumstances is differing from conventional websites social network is considered in this work. A model of dynamic rumor influence reduction with classification (MDRIPC) is projected in this work. Major objective of this paper is to reduce the impact of the rumor from the dataset which is actually amount of users admitted and sent rumor. This is achieved using the classifiers such as Naive Bayes, Random forest and KNN by blocking an exact set of nodes with imbalanced dataset. But some records belonging to same group are considerably in huge number and some are very uncommon and hence several datasets are imbalanced. Also the performance of a classifier is influenced significantly by this imbalanced nature of the datasets. Random sampling technique is applied to handle with this issue. Semi Supervised Clustering Algorithm (SSCA) is employed furthermore, to solve rumor propagation issue and this is implemented through the information gathered by the analysis of social networks. Supported large-scale world networks are utilized for the experiments and the effectiveness of the three classifiers are validated.

In this paper, we introduce a neural network framework for semi-supervised clustering (SSC) with pairwise (must-link or cannot-link) constraints. In contrast to existing approaches, we decompose SSC into two simpler classification tasks/stages: the first stage uses a pair of Siamese neural networks to label the unlabeled pairs of points as must-link or cannot-link; the second stage uses the fully pairwise-labeled dataset produced by the first stage in a supervised neural-network-based clustering method. The proposed approach, S3C2 (Semi-Supervised Siamese Classifiers for Clustering), is motivated by the observation that binary classification (such as assigning pairwise relations) is usually easier than multi-class clustering with partial supervision. On the other hand, being classification-based, our method solves only well-defined classification problems, rather than less well specified clustering tasks. Extensive experiments on various datasets demonstrate the high performance of the proposed method.

We consider the problem of data clustering with unidentified feature quality and when a small amount of labelled data is provided. An unsupervised sparse clustering method can be employed in order to detect the subgroup of features necessary for clustering and a semi-supervised method can use the labelled data to create constraints and enhance the clustering solution. In this paper we propose a K-Means variant that employs these techniques. We show that the algorithm maintains the high performance of other semi-supervised algorithms and in addition preserves the ability to identify informative from uninformative features. We examine the performance of the algorithm on synthetic and real world data sets. We use scenarios of different number and types of constraints as well as different clustering initialisation methods., Comment: Under consideration at Pattern Recognition Letters. 2 figures, 1 table, supplementary material

Impact Factor: 8.038; International audience; Semi-supervised consensus clustering, also called semi-supervised ensemble clustering, is a recently emerged technique that integrates prior knowledge into consensus clustering in order to improve the quality of the clustering result. In this article, we propose a novel semi-supervised consensus clustering algorithm extending the previous work on the MultiCons multiple consensus clustering approach. By using closed pattern mining technique, the proposed Semi-MultiCons algorithm manages to generate a recommended consensus solution with a relevant inferred number of clusters k based on ensemble members with different k and pairwise constraints. Compared with other semi-supervised and/or consensus clustering approaches, Semi-MultiCons does not require the number of generated clusters k as an input parameter, and is able to alleviate the widely reported negative effect related to the integration of constraints into clustering. The experimental results demonstrate that the proposed method outperforms state of the art semi-supervised consensus clustering algorithms.

Анализ методов решения задач кластеризации с частичным привлечением учителя (Constrained Clustering или Semi Supervised Clustering) показал, что большинство из них являются приближенными. Точные методы технологии программирования в ограничениях позволяют решать задачи комбинаторного поиска, характеризующиеся большой размерностью. Однако, данные средства всё ещё редко применяются в задачах Constrained Clustering. Сделан вывод о перспективности разработки новых методов удовлетворения ограничений для различных постановок задач Constrained Clustering., An analysis of the methods of solving Semi Supervised Clustering(Constrained Clustering) problems showed that most of them are approximate. Exact methods of constraint programming technology allow solving combinatorial search problems of large dimension. However, these tools are still rarely used in Constrained Clustering problems. The conclusion is drawn on the prospects of developing new constraint satisfaction methods for various statements of Constrained Clustering problem., №9(9) (2020)

The main purpose of the intrusion detection system (IDS) is to detect a network attack and respond to the network intrusion. Existing supervised IDSs require a large amount of tag data as the training data, and there is almost no effect on the unknown attacks. Traditional unsupervised intrusion systems have problems including low accuracy and the inability to provide specific information regarding the detected attacks. To solve the above problems, we propose a multilayer IDS based on semi-supervised clustering. This system solves the problem of insufficient training data by using tag extension technology and genetic algorithm, and solves the problem of unsupervised clustering unable to provide specific information of attack by using the idea of semi-supervised clustering. We use the NSL-KDD dataset to conduct the experiments. The simulation results show that the proposed IDS only needs a small amount of training data to obtain better performance, especially for lower frequency attacks.

Attributable to the fact that the clustering accuracy is not high under the original Affinity Propagation (AP) algorithm, that is greatly affected by preference (P) when adjusting it to obtain the true class number of clustering, this paper proposed an improved AP algorithm named Semi-supervised Hierarchical Optimization-based AP algorithm (SHO-AP). The algorithm introduces the idea of semi-supervision, by setting a certain proportion of label data and using the AP to cluster, then establish the supervision and non-supervision information matrix to optimize, and combine the result of AP algorithm. We utilize hierarchical optimization to combine the final clustering results. The result of the experiment on UCI data-sets shows that the proposed algorithm achieves higher quality than the traditional AP algorithm and the number of classes is much closer to the real number.

Cross-company defect prediction (CCDP) is a practical way that trains a prediction model by exploiting one or multiple projects of a source company and then applies the model to a target company. Unfortunately, larger irrelevant cross-company (CC) data usually make it difficult to build a prediction model with high performance. On the other hand, brute force leveraging of CC data poorly related to within-company data may decrease the prediction model performance. To address such issues, we aim to provide an effective solution for CCDP. First, we propose a novel semi-supervised clustering-based data filtering method (i.e., SSDBSCAN filter) to filter out irrelevant CC data. Second, based on the filtered CC data, we for the first time introduce multi-source TrAdaBoost algorithm, an effective transfer learning method, into CCDP to import knowledge not from one but from multiple sources to avoid negative transfer. Experiments on 15 public datasets indicate that: (1) our proposed SSDBSCAN filter achieves better overall performance than compared data filtering methods; (2) our proposed CCDP approach achieves the best overall performance among all tested CCDP approaches; and (3) our proposed CCDP approach performs significantly better than with-company defect prediction models.

In the current trend, people pay attention to healthcare services more and more. Liver-related diseases, same as other common diseases, have many effects on human health. Herein, we introduce a novel model to liver disease diagnosis problem based on the hybrid of fuzzy min–max neural network, semi-supervised clustering, and Genetic Algorithm. Our model is evaluated on four popular datasets from the University of California Irvine and one real dataset. Our proposed model has a better performance and much smaller number of hyper boxes compared with other available methods.

Nowadays, data mining becomes a very important research filed for knowledge discovery process. Among various data mining techniques, we focus on studying how a small amount of prior knowledge can be effectively incorporated into some popular clustering techniques, not only to improve the existing models, but also to develop novel semi-supervised clustering methods, especially for the categorization of high dimensional text documents. To be more specific, our objective is to investigate into some of the key performance characteristics of a good semi-supervised clustering method, and make them achievable in our purposed methods, such as: how to effectively incorporate the knowledge to guide the cluster search, accurately capture the underlying structure of the data, and achieve high capability of handling overlaps etc. In other words, our final goal is to develop some simple, fast and highly applicable clustering methods which aim to achieve good results with high quality and effectiveness with the help of the prior knowledge. Fuzzy co-clustering (FCC) is a type of clustering approaches that has shown its capability for handling high dimensional textual data categorization by simultaneously grouping the documents and words into some co-clusters. Meanwhile, for a document which intuitively spans multiple topics, FCC is also able to capture the degree of memberships of that document to each topic. Under the FCC framework, we proposed three different semi-supervised approaches, namely Semi-Supervised Fuzzy Co-clustering with Labelling (SS-FCL), Semi-Supervised Fuzzy Co-clustering with Constraints (SS-FCC) and Dual Semi-Supervised Heuristic Fuzzy Co-clustering with Ruspini’s condition (DSS-HFCR), respectively. Two types of prior knowledge in the forms of class labels and pair-wise constraints from the document domain are incorporated into SS-FCL and SS-FCC through different additional supervised constraint terms. Other than the categorization results of the documents, these two approaches also generate a group of work ranking clusters, which are useful in other data mining techniques, such as text summarization. Meanwhile, a heuristic dual-partitioning based approach called DSS-HFCR is also proposed in order to make full use of the available prior knowledge in terms of pair-wise constraints from both document and word domain. Moreover, DSS-HFCR can be directly downgraded to a simplified version if the prior knowledge is available from only a single domain Through extensive experimental study on a number of benchmark textual datasets, we demonstrate how these approaches make good use of the knowledge to guide the cluster search during the clustering process for a better performance in terms of both accuracy and efficiency. Some useful guidelines for parameter selection are also discussed. A case study of sentiment data analysis by applying DSS-HFCR demonstrates the strength of our proposed methods in the specific application area. FCC model is suitable for handling large sparse textual datasets, as it avoids applying an explicit similarity measure between two documents. However, similarity measure is still one of the most essential factors in many discriminative clustering approaches, and most of the these approaches still make use of only a single reference point (viewpoint) i.e. the origin for the similarity assessment. It is interesting and challenging to explore more effective similarity measures for high dimensional textual data, especially when some prior knowledge is available to the user. For the second part of the thesis, inspired by a recently proposed multi-viewpoint based similarity measure (MVS) [1], we introduce another novel semi-supervised clustering framework, which is able to utilize multiple appropriate viewpoints for a more informative and effective similarity assessment by incorporating two types of knowledge. With the help of a small number of class labels or pair-wise constraints in the dataset, we formulate two MVS measures, and subsequently propose two new MVS-based clustering approaches: Label-based Clustering with Multi-Viewpoint based Similarity (namely LMVS) and Pair-wise Constraints-based Clustering with Multi-Viewpoint based Similarity (namely PMVS). Comparing with the existing semi-supervised clustering techniques, the key strength of LMVS and PMVS is a more effective similarity measure can be directly formulated in the MVS manner with the help of the knowledge, and immediately applied to clustering, rather than learned by an independent distance metric learning process before the real clustering process is carried out. Some validity tests are conducted to show the strength of the measures, and systematical theoretical analysis is also provided to explain how the prior knowledge is utilized for both similarity enhancement and search-guiding purpose during the clustering process. At the same time, some potential issues of MVSC reported in [1] can be successfully addressed. At last, extensive experimental study on a large number of benchmark textual datasets are presented to demonstrate the effectiveness and verify the merit of LMVS & PMVS Clustering, compared with other start-of-the-art semi-supervised clustering/learning approaches. DOCTOR OF PHILOSOPHY (EEE)

International audience; In semi-supervised graph clustering setting, an expert provides cluster membership of few nodes. This little amount of information allows one to achieve high accuracy clustering using efficient computational procedures. Our main goal is to provide a theoretical justification why the graph-based semi-supervised learning works very well. Specifically, for the Stochastic Block Model in the moderately sparse regime, we prove that popular semi-supervised clustering methods like Label Spreading achieve asymptotically almost exact recovery as long as the fraction of labeled nodes does not go to zero and the average degree goes to infinity.

Semi-supervised learning is drawing increasing attention in the era of big data, as the gap between the abundance of cheap, automatically collected unlabeled data and the scarcity of labeled data that are laborious and expensive to obtain is dramatically increasing. In this thesis, we first introduce a unified view of density-based clustering algorithms. Then, we build upon this view and bridge the areas of semi-supervised clustering and classification under a common umbrella of density-based techniques. We show that there are close relations between density-based clustering algorithms and the graph-based approach for transductive classification. These relations are then used as a basis for a new framework for semi-supervised classification based on building-blocks from density-based clustering. This framework is not only efficient and effective, but it is also statistically sound. We also generalize the core algorithm of the framework HDBSCAN* so that it can also perform semi-supervised clustering by directly taking advantage of any fraction of labeled data that may be available, rather than instance-level pairwise constraints. Experimental results on a large collection of datasets show the advantages of the proposed approach both for semi-supervised classification, as well as for semi-supervised clustering. In addition, we evaluate the semi-supervised learning algorithms to determine relationships between chemical structure and biological activity in datasets from Medicinal Chemistry. The datasets evaluated in this area are characterized by a low number of labeled examples, a high dimensionality, and in some cases, do not have a clear relationship between chemical structure and biological activity, which makes it difficult to use classification techniques and analyze chemical phenomena. We implement and validate semi-supervised classification approaches that are appropriate for data analysis in Medicinal Chemistry. O aprendizado semissupervisionado obteve, no decorrer do tempo, cada vez mais significância na área de big data, uma vez que a lacuna entre a abundância de dados não rotulados coletados de forma rápida e barada e a escassez de dados rotulados que são trabalhosos e caros de obter, está aumentando drasticamente. No presente documento, apresentamos uma visão unificada dos algoritmos de agrupamento baseados em densidade. Depois, utilizamos essa visão unificada para construção de uma ponte entre as áreas de agrupamento e classificação semissupervisionadas. Mostramos que existem relações entre algoritmos de agrupamento baseados em densidade e a abordagem baseada em grafos para classificação transdutiva. Essas relações são então usadas como base para um novo framework de classificação semissupervisionada com base em algoritmos baseados em densidade. Assim, definimos um novo framework que pode ser dividido em pequenos blocos de construção, com base nos algoritmos baseado em densidade. Essa estrutura é eficiente, eficaz e também estatisticamente sólida. Também generalizamos o framework HDBSCAN* para que ele também pudesse realizar agrupamento semissupervisionado utilizando diretamente qualquer fração de dados rotulados que possam estar disponíveis, ao invés de um conjunto de restrições. Resultados experimentais em uma grande coleção de base de dados mostram as vantagens da abordagem proposta tanto para classificação semissupervisionada quanto para o agrupamento semissupervisionado. Além disso, avaliamos os algoritmos de aprendizado semissupervisionados para determinar as relações entre a estrutura química e a atividade biológica em conjuntos de dados da área de Química Medicinal. Os conjuntos de dados avaliados nesta área são caracterizados por um baixo número de exemplos rotulados, alta dimensionalidade e, em alguns casos, não possuem uma relação clara entre estrutura química e atividade biológica, o que dificulta a aplicação de técnicas aprendizado supervisionado. Após implementação e validação das abordagens semissupervisionadas, verificamos que os mesmos podem ser uma boa opção para análise de dados em Química Medicinal.

In this paper, we consider the semi-supervised clustering problem, where the prior knowledge is formalized as the Cannot-Link (CL) and Must-Link (ML) pairwise constraints. We propose an algorithm called SemiSync that tackles this problem from a novel perspective: synchronization. The basic idea is to regard the data points as a set of (constrained) phase oscillators, and simulate their dynamics to form clusters automatically. SemiSync allows dynamically propagating the constraints to unlabelled data points driven by their local data distributions, which effectively boosts the clustering performance even if little prior knowledge is available. We experimentally demonstrate the effectiveness of the proposed method.

Semi-supervised clustering has attracted lots of reserach interest due to its broad applications, and many methods have been presented. However there is still much space for improvement, (1) How to learn more discriminative feature representations to assist the traditional clustering methods; (2) How to make use of both the labeled and unlabelled data simultaneously and effectively during the process of clustering. To address these issues, we propose a novel semi-supervised clustering based on deep metric learning, namely SSCDML. By leveraging deep metric learning and semi-supervised learning effectively in a novel way, SSCDML dynamically update the unlabelled to labeled data through the limited labeled samples and obtain more meaningful data features, which make the classifier model more robust and the clustering results more accurate. Experimental results on Mnist, YaleB, and 20 Newsgroups databases demonstrate the high effectiveness of our proposed approach.

The recognition process of 3D face in cloud environment is vulnerable to the interference of external environment, resulting in poor recognition accuracy, therefore, a 3D face recognition method in cloud environment based on semi supervised clustering algorithm is proposed in this paper, after Harris feature points of successful matching of a 2D image are mapped into 3D space, surface fitting method estimated by the least squares is used to extract the curvature information corresponding to feature points, maximum and minimum principal curvatures are constructed as the final curvature eigenvector. The semi supervised clustering algorithm is introduced to perform the cluster judgement to decide if the sample is labeled 3D face sample. According to the class probability or membership degree of 3D face samples to preselect partial samples, labeled 3D face samples training classifier is established, after the second-selection is processed for the preselected 3D face samples according to the classification confidence, the selected results and the predicted labels are added into labeled samples. The iterations are applied to the semi supervised clustering algorithm to cluster the original labeled samples, new labeled samples and label the rest of the unlabeled samples,. The iterative process is repeated until all samples have been marked, and the labeled samples are identified 3D face data. The experiment combined proposed method with cloud computing platform, the experimental results show that the proposed method has high recognition accuracy and efficiency.

The label propagation algorithm is a well-known semi-supervised clustering method, which uses pre-given partial labels as constraints to predict the labels of unlabeled data. However, the algorithm has the following limitations: (1) it does not fully consider the misalignment between the pre-given labels and clustering labels, and (2) it only uses label information as clustering constraints. Real applications not only contain partial label information but pairwise constraints on a dataset. To overcome these deficiencies, a new version of the label propagation algorithm is proposed, which makes use of pairwise relations of labels as constraints to construct an optimization model for spreading labels. Experimental analysis was used to compare the proposed algorithm with 8 other semi-supervised clustering algorithms on 11 benchmark datasets. The experimental results demonstrated that the proposed algorithm is more effective than other algorithms.

As soft computing extensions of hard C-means (HCM) clustering, rough C-means (RCM) and rough set C-means (RSCM), which can deal with positive and possible cluster memberships based on rough set theory, have been proposed and utilized for detecting vague boundaries among clusters. Semi-supervised clustering schemes that utilize information of partial labeled objects are promising approaches for improving the classification performance. In this study, we consider how to introduce semi-supervised clustering schemes to RCM and RSCM. Furthermore, we confirm the effectiveness of the proposed methods through numerical experiments.

In the proposed work, our research focus is on semi-supervised clustering, which uses a small amount of supervised data in the form of class labels or pairwise constraints on some examples to aid unsupervised clustering. In addition, we have also included incomplete data with missing features. In the proposed clustering algorithm, we have embedded the optimal completion strategy (OCS) in semi-supervised clustering algorithm which imputes the missing values by the maximum likelihood estimate in an iterative optimization procedure. A series of detailed experimental analysis on real data sets is done to illustrate the main ideas discussed in the study work.

One of the least invasive means of diagnosing certain diseases is through the use of imaging tests. Automated analysis usually begins by detecting the regions to be investigated and extracting relevant information such as shape and texture. In this paper, we propose a semi-supervised clustering algorithm called Seeded Fuzzy C-means to segment regions of interest in medical images. Seeded Fuzzy C-means is based on the well-known Fuzzy C-means; however, it also uses information provided by the physician to insert constraints during group choices. In this way, the element is placed in a group with a higher degree of certainty. We evaluated the proposed algorithm on a total of 2,200 images in leukemia, skin cancer, cervical cancer, and glaucoma image databases. Two semi-supervised clustering algorithms were implemented from literature to perform this analysis. The results illustrate the ability of the algorithm to efficiently assist physicians in detecting regions of interest, as it achieved an “Excellent” Kappa index in most of the tests.

The amygdala plays an important role in emotional functions and its dysfunction is considered to be associated with multiple psychiatric disorders in humans. Cytoarchitectonic mapping has demonstrated that the human amygdala complex comprises several subregions. However, it's difficult to delineate boundaries of these subregions in vivo even if using state of the art high resolution structural MRI. Previous attempts to parcellate this small structure using unsupervised clustering methods based on resting state fMRI data suffered from the low spatial resolution of typical fMRI data, and it remains challenging for the unsupervised methods to define subregions of the amygdala in vivo. In this study, we developed a novel brain parcellation method to segment the human amygdala into spatially contiguous subregions based on 7T high resolution fMRI data. The parcellation was implemented using a semi-supervised spectral clustering (SSC) algorithm at an individual subject level. Under guidance of prior information derived from the Julich cytoarchitectonic atlas, our method clustered voxels of the amygdala into subregions according to similarity measures of their functional signals. As a result, three distinct amygdala subregions can be obtained in each hemisphere for every individual subject. Compared with the cytoarchitectonic atlas, our method achieved better performance in terms of subregional functional homogeneity. Validation experiments have also demonstrated that the amygdala subregions obtained by our method have distinctive, lateralized functional connectivity (FC) patterns. Our study has demonstrated that the semi-supervised brain parcellation method is a powerful tool for exploring amygdala subregional functions.

Density-based clustering is an important class of approaches to data clustering due to good performance. Among this class of approaches, DenPeak is an effective density-based clustering method that can automatically find the number of clusters and find arbitrary-shape clusters in relative easy scenarios. However, in many situations, it is usually hard for DenPeak to find an appropriate number of clusters without supervision or prior knowledge. In addition, DenPeak often fails to find local structures of each cluster since it assigns only one center to each cluster. To address these problems, we introduce a novel semi-supervised DenPeak clustering (SSDC) method by introducing pairwise constraints or side information to guide the cluster process. These pairwise constraints or side information improve the clustering performance by explicitly indicating the affiliated cluster of data samples in each pair. Concretely, SSDC firstly generates a relatively large number of temporary clusters, and then merges them with the assistance of samples’ pairwise constraints and temporary clusters’ adjacent information. The proposed SSDC can significantly improve the performance of DenPeak. Its superiority to state-of-the-art clustering methods has been empirically demonstrated on both artificial and real data sets.

In this paper, we introduce a novel interactive framework to handle both instance-level and temporal smoothness constraints for clustering large temporal data. It consists of a constrained clustering algorithm, called CVQE+, which optimizes the clustering quality, constraint violation and the historical cost between consecutive data snapshots. At the center of our framework is a simple yet effective active learning technique, named Border, for iteratively selecting the most informative pairs of objects to query users about, and updating the clustering with new constraints. Those constraints are then propagated inside each data snapshot and between snapshots via two schemes, called constraint inheritance and constraint propagation, to further enhance the results. Experiments show better or comparable clustering results than state-of-the-art techniques as well as high scalability for large datasets.

With the rapid growth of mashup resources, clustering mashup services according to the functions of the mashup services has become an effective way to improve the quality of mashup services management. Clustering is a learning task that classifies individuals or objects into different clusters based on the similarity. The purpose of clustering is to maximize the homogeneity of elements in the same cluster and maximize the heterogeneity of the elements in different clusters. It is a multivariate statistical method for classification. However, compared with the supervised classification, the clustering’s ability to categorize is much weaker. Existing methods for mashup services clustering mostly focus on utilizing key features from WSDL documents directly. In this paper, we proposed a method to improve the categorize ability of clustering. That is, applying supervised thought to cluster mashup services. First, taking basic clustering operations on the WSDL documents of mashups to obtain the clustering result for each element. Then, using the WSDL documents as training data, and the clustering results from the first step as pseudo-tags to train a classification learner. Finally, classifying mashups with this classification learner to get the final clustering results.

In this paper, we propose a real data clustering method based on active learning. Clustering methods are difficult to apply to real data for two reasons. First, real data may include outliers that adversely affect clustering. Second, the clustering parameters such as the number of clusters cannot be made constant because the number of classes of real data may increase as time goes by. To solve the first problem, we focus on labeling outliers. Therefore, we develop a stream-based active learning framework for clustering. The active learning framework enables us to label the outliers intensively. To solve the second problem, we also develop an algorithm to automatically set clustering parameters. This algorithm can automatically set the clustering parameters with some labeled samples. The experimental results show that our method can deal with the problems mentioned above better than the conventional clustering methods.

In this paper, we propose a novel system for capsule endoscopy (CE) summarisation that has two main components. The first component consists of the Semi-Supervised Clustering and Local Scale Learning (SS-LSL) algorithm. This algorithm is used to group video frames into prototypical clusters that summarise the CE video. The constraints consist of pairs of frames that should not be included in the same cluster. These constraints are deduced from the training frames to help in guiding the clustering process. The second component of the system consists of a novel relational motion histogram descriptor that is designed to represent the local motion distribution between two contiguous frames. The main idea is to identify “highlight” frames which contain typical variations within the frame collection. These variations are due to different pathologies, small tumours and other subtle abnormalities of the small intestine and so on. SS-LSL algorithm is assessed using synthetic data sets, and proved to outperform sim...