Maximum Clustering Research Articles

Unsupervised learning by minimal entropy encoding.

Following basic principles of information-theoretic learning, in this paper, we propose a novel approach to data clustering, referred to as minimal entropy encoding (MEE), which is based on a set of functions (features) projecting each input onto a minimum entropy configuration (code). Inspired by traditional parsimony principles, we seek solutions in reproducing kernel Hilbert spaces and then we prove that the encoding functions are expressed in terms of kernel expansion. In order to avoid trivial solutions, the developed features must be as different as possible by means of a soft constraint on the empirical estimation of the entropy associated with the encoding functions. This leads to an unconstrained optimization problem that can be efficiently solved by conjugate gradient. We also investigate an optimization strategy based on concave-convex algorithms. The relationships with maximum margin clustering are studied, showing that MEE overcomes some of its critical issues, such as the lack of a multiclass extension and the need to face problems with a large number of constraints. A massive evaluation on several benchmarks of the proposed approach shows improvements over state-of-the-art techniques, both in terms of accuracy and computational complexity.

Identification of Hierarchical and Overlapping Functional Modules in PPI Networks

Various evidences have demonstrated that functional modules are overlapping and hierarchically organized in protein-protein interaction (PPI) networks. Up to now, few methods are able to identify both overlapping and hierarchical functional modules in PPI networks. In this paper, a new hierarchical clustering algorithm, called OH-PIN, is proposed based on the overlapping M_clusters, λ-module, and a new concept of clustering coefficient between two clusters. By recursively merging two clusters with the maximum clustering coefficient, OH-PIN finally assembles all M_clusters into λ -modules. Since M_clusters are overlapping, λ -modules based on them are also overlapping. Thus, OH-PIN can detect a hierarchical organization of overlapping modules by tuning the value of λ. The hierarchical organization is similar to the hierarchical organization of GO annotations and that of the known complexes in MIPS. To compare the performance of OH-PIN and other existing competing algorithms, we apply them to the yeast PPI network. The experimental results show that OH-PIN outperforms the existing algorithms in terms of the functional enrichment and matching with known protein complexes.

Semi-Supervised Maximum Margin Clustering with Pairwise Constraints

The pairwise constraints specifying whether a pair of samples should be grouped together or not have been successfully incorporated into the conventional clustering methods such as k-means and spectral clustering for the performance enhancement. Nevertheless, the issue of pairwise constraints has not been well studied in the recently proposed maximum margin clustering (MMC), which extends the maximum margin framework in supervised learning for clustering and often shows a promising performance. This paper therefore proposes a pairwise constrained MMC algorithm. Based on the maximum margin idea in MMC, we propose a set of effective loss functions for discouraging the violation of given pairwise constraints. For the resulting optimization problem, we show that the original nonconvex problem in our approach can be decomposed into a sequence of convex quadratic program problems via constrained concave-convex procedure (CCCP). Subsequently, we present an efficient subgradient projection optimization method to solve each convex problem in the CCCP sequence. Experiments on a number of real-world data sets show that the proposed constrained MMC algorithm is scalable and outperforms the existing constrained MMC approach as well as the typical semi-supervised clustering counterparts.

Identification of differentially expressed genes in gastric cancer by high density cDNA microarray

The identification of molecular markers for diagnosis, treatment, and prognosis is a significant issue in the management of patients with gastric cancer. We compared the expression profiles of 23 gastric cancers and 22 normal gastric tissues using cDNA microarrays. We divided the samples into two sets, 11 pairs as a training set and 12 unpaired gastric cancer and 11 unpaired normal gastric tissues as a test set. We selected significant genes in the training set and validated the significance of the genes in the test set. We obtained 238 classifier genes that showed a maximum cross-validation probability and clear hierarchical clustering pattern in the training set, and showed excellent class prediction probability in the independent test set. The classifier genes consisted of known genes related to the biological features of cancer and 28% unknown genes. We obtained genome-wide molecular signatures of gastric cancer, which provides preliminary exploration data for the pathophysiology of gastric cancer.

Clustering Very Low-Income, Insured Women’s Mammography Screening Barriers Into Potentially Functional Subgroups

BackgroundMammography screening is essential for early detection of breast cancer and increased survival rates. Women, particularly those of low socioeconomic status, face barriers that impede their screening adherence. Although many studies have sought to identify these barriers, more research is needed on to address these obstacles in practice. The objective of this study is to divide mammography screening barriers into functional clusters using empirical evidence, which may guide the development of effective mammography screening promotion messages. MethodsA sample of 173 low-income White and Black women randomly selected from a managed care organization rated each of 21 potential mammography barriers on a scale ranging from “strongly agree” to “strongly disagree.” A maximum clustering similarity method was used to identify relevant clusters of screening barriers. ResultsFive clusters were derived, with a high similarity index (0.881). Each cluster was named to reflect the shared theme of the barriers within it: Perceived Lack of Value in Health Care, Lack of Information, Mistrust/Skepticism, Medical Delay Behavior, and Anxiety/No Control. A dominant barrier within each cluster was identified, and bivariate correlation coefficients were reported. ConclusionCluster analysis yielded five distinct subgroups of mammography screening barriers.

Three-dimensional Lagrangian Voronoï analysis for clustering of particles and bubbles in turbulence

AbstractThree-dimensional Voronoï analysis is used to quantify the clustering of inertial particles in homogeneous isotropic turbulence using data sets from numerics in the point particle limit and one experimental data set. We study the clustering behaviour at different density ratios, particle response times (i.e. Stokes numbers $\mathit{St}$) and two Taylor–Reynolds numbers (${\mathit{Re}}_{\lambda } = 75$ and 180). The probability density functions (p.d.f.s) of the Voronoï cell volumes of light and heavy particles show different behaviour from that of randomly distributed particles, i.e. fluid tracers, implying that clustering is present. The standard deviation of the p.d.f. normalized by that of randomly distributed particles is used to quantify the clustering. The clustering for both light and heavy particles is stronger for higher ${\mathit{Re}}_{\lambda } $. Light particles show maximum clustering for $\mathit{St}$ around 1–2 for both Taylor–Reynolds numbers. The experimental data set shows reasonable agreement with the numerical results. The results are consistent with previous investigations employing other approaches to quantify the clustering. We also present the joint p.d.f.s of enstrophy and Voronoï volumes and their Lagrangian autocorrelations. The small Voronoï volumes of light particles correspond to regions of higher enstrophy than those of heavy particles, indicating that light particles cluster in higher vorticity regions. The Lagrangian temporal autocorrelation function of Voronoï volumes shows that the clustering of light particles lasts much longer than that of heavy or neutrally buoyant particles. Due to inertial effects arising from the density contrast with the surrounding liquid, light and heavy particles remain clustered for much longer times than the flow structures which cause the clustering.

Research history generation from metainformation of research papers using maximum margin clustering

Recently, analysing research papers to understand research trends researcher's research topics automatically from metainformation of research papers published on the internet. Our method is based on Maximum Margin Clustering (MMC). We describe how to represent research papers in form of vectors using metainformation about them and how to initialise the hyperplane for MMC automatically. In the experiments, we show that the purity of our method is higher than that achieved in previous work based on k-Means (0.58 vs 0.35) and entropy of our method is lower than that of previous work (0.415 vs 0.47). Experiment results also illustrates that keyword information of research papers affects the most to clustering result.

Multiple-Region Segmentation Without Supervision by Adaptive Global Maximum Clustering

In this paper, we propose a new method of segmenting an image into several sets of pixels with similar intensity values called regions. A multiple-region segmentation problem is unstable because the result considerably depends on the number of regions given a priori. Therefore, one of the most important tasks in solving the problem is automatically finding the number of regions. The method we propose is able to find the reasonable number of distinct regions not only for clean images but also for noisy ones. Our method is made up of two procedures. First, we develop the adaptive global maximum clustering. In this procedure, we deal with an image histogram and automatically obtain the number of significant local maxima of the histogram. This number indicates the number of different regions in the image. Second, we derive a simple and fast calculation to segment an image composed of distinct multiple regions. Then, we split an image into multiple regions according to the previous procedure. Finally, we show the efficiency of our method by comparing it with other previous methods.

Origin of Microstructural Irreversibility in Ni-Ti Based Shape Memory Alloys during Thermal Cycling

Different microstructures of Ni-Ti- and Ni-Ti-Fe-based shape memory alloys were subjected to thermal cycling: dipping in liquid nitrogen, for approximately 5 minutes, and then bringing it back to room temperature or austenite (cubic: B2) ↔ martensite (monoclinic: B19′) reversible solid-state phase transformation. Direct electron backscattered diffraction (EBSD) observations could bring out aspects of microstructural irreversibilities: namely, changes in grain size, misorientation buildup, and presence of retained martensite. The average changes in grain size (Δd) differed by almost 2 to 4 times between different microstructures. The highest Δd was typically observed in structures having maximum clustering of fine (d < 5 μm) grains. The sample with highest Δd was also subjected to multiple thermal cycling. Although Δd scaled linearly with d after the first thermal cycle, the scatter increased during subsequent thermal cycles. Grain or orientations deviating from the linear behavior were clearly anisotropic crystallographically. With repeated thermal cycling, the patterns of changes in Δd, austenite misorientation, and retained martensite content were similar. A phenomenological model or hypothesis, based on 40 deg \( \left\langle {001} \right\rangle \) orientation relationship between austenite and martensite phases, was proposed to address the observed patterns of microstructural irreversibility.

Genetic diversity and population structure of Indonesian native chickens based on single nucleotide polymorphism markers

Indonesian native chickens are considered an important genetic resource, particularly with respect to their excellent traits for meat and egg production. However, few molecular genetic studies of these native chickens have been conducted. We analyzed the genetic diversity and differentiation of 4 populations of Indonesian native chickens: Black Kedu (BK), Kedu (KD), Kampung (LOC), and Arab (AR). Blood samples from 188 individuals were collected in central and western Java. Genomic DNA was genotyped using 98 autosomal SNP markers, of which 87 were found to be polymorphic. The proportion of polymorphic loci and the average heterozygosity of each population were in the range of 0.765 to 0.878 and 0.224 to 0.263, respectively. The 4 populations of Indonesian chickens appeared to be derived from 3 genetic populations (K = 3): maximum likelihood clustering showed that the BK variety and AR breed were each assigned to a distinct cluster, whereas the LOC ecotype and KD variety were admixed populations with similar proportions of membership. Principal components analysis revealed that eigenvector 1 separated BK and AR from the other 2 populations. Neighbor-joining trees constructed from pairwise distance matrix (F(ST)) estimates, for individuals and between populations, corroborated that the LOC ecotype and KD variety were related closely, whereas the BK variety and AR breed diverged at greater distances. These results also confirmed the usefulness of SNP markers for the study of genetic diversity.

On first quartile journals which are not of highest impact

Here we study the relationship between journal quartile rankings of ISI impact factor (at the 2010) and journal classification in four impact classes, i.e., highest impact, medium highest impact, medium lowest impact, and lowest impact journals in subject category computer science artificial intelligence. To this aim, we use fuzzy maximum likelihood estimation clustering in order to identify groups of journals sharing similar characteristics in a multivariate indicator space. The seven variables used in this analysis are: (1) Scimago Journal Ranking (SJR); (2) H-Index (H); (3) ISI impact factor (IF); (4) 5-Year Impact Factor (5IF); (5) Immediacy Index (II); (6) Eigenfactor Score (ES); and (7) Article Influence Score (AIS). The fuzzy clustering allows impact classes to overlap, thereby accommodating for uncertainty related to the confusion about the impact class attribution for a journal and vagueness in impact classes definition. This paper demonstrates the complex relationship between quartiles of ISI impact factor and journal impact classes in the multivariate indicator space. And that several indicators should be used for a distinct analysis of structural changes at the score distribution of journals in a subject category. Here we propose it can be performed in a multivariate indicator space using a fuzzy classifier.

Group and Region Based Parallel Compression Method Using Signal Subspace Projection and Band Clustering for Hyperspectral Imagery

In this study, a novel group and region based parallel compression approach is proposed for hyperspectral imagery. The proposed approach contains two algorithms, which are clustering signal subspace projection (CSSP) and the maximum correlation band clustering (MCBC). The CSSP first divides the image into proper regions by transforming the high dimensional image data into one dimensional projection length. The MCBC partitions the spectral bands into several groups according to their associated band correlation for each image region. The image data with high degree correlations in spatial/spectral domains are then gathered in groups. Then, the grouped image data is further compressed by Principal Components Analysis (PCA)-based spectral/spatial hyper-spectral image compression techniques. Furthermore, to accelerate the computing efficiency, we present a parallel architecture of the proposed compression approach by using parallel cluster computing techniques. Simulation results performed on AVIRIS images have shown that the proposed group and region based approach performs better than standard 3D hyperspectral image compression. Moreover, the proposed approach achieves better computation efficiency than the direct combination of PCA and JPEG2000 under the same compression ratio.

Maximum Margin Multiple Instance Clustering With Applications to Image and Text Clustering

In multiple instance learning problems, patterns are often given as bags and each bag consists of some instances. Most of existing research in the area focuses on multiple instance classification and multiple instance regression, while very limited work has been conducted for multiple instance clustering (MIC). This paper formulates a novel framework, maximum margin multiple instance clustering (M(3)IC), for MIC. However, it is impractical to directly solve the optimization problem of M(3)IC. Therefore, M(3)IC is relaxed in this paper to enable an efficient optimization solution with a combination of the constrained concave-convex procedure and the cutting plane method. Furthermore, this paper presents some important properties of the proposed method and discusses the relationship between the proposed method and some other related ones. An extensive set of empirical results are shown to demonstrate the advantages of the proposed method against existing research for both effectiveness and efficiency.

Improved Estimate-to-Track Association Method for Track Continuity

The probability hypothesis density (PHD) filter is a practical alternative to the optimal Bayesian multi-target filter based on finite set statistics. However, the PHD filter keeps no record of target identities and hence does not produce track-valued estimates of individual targets. To solve this problem, an improved estimate-to-track association method for track continuity is proposed in this paper. Firstly, a multi-step prediction of current target states is made, and then the weighted labels are assigned to them according to the inertia. Secondly, the fuzzy membership degrees of the current state estimates belonging to the tracks are obtained by utilizing the maximum entropy fuzzy clustering. Finally, the tracks are maintained by integrating all this information. Different from the traditional estimate-to-track association method, the proposed algorithm does not update the track information by simply summing the log likelihood ratios between adjacent scans, but take the entire multi-scan information into account by the operations such as weighting and clustering. The simulation results show that the proposed algorithm can maintain target tracks more accurately, even when the targets cross each other, implying strong robustness and excellent performance of track continuity.

Maximum clustering coefficient of graphs with given number of vertices and edges

The clustering coefficient is one of the most important quantities characterizing complex networks. It is often said that many networks in the real world have high clustering coefficients. However, properties of the clustering coefficient itself have not been discussed much in the literature. In this paper, some fundamental properties of the clustering coefficient are studied. First we try to find the maximum value of the clustering coefficient of graphs with the given number of vertices and edges. Next we present some classes of graphs such that each member maximizes the clustering coefficient among its neighbors having the same number of vertices and edges.

MCS: A Method for Finding the Number of Clusters

This paper proposes a maximum clustering similarity (MCS) method for determining the number of clusters in a data set by studying the behavior of similarity indices comparing two (of several) clustering methods. The similarity between the two clusterings is calculated at the same number of clusters, using the indices of Rand (R), Fowlkes and Mallows (FM), and Kulczynski (K) each corrected for chance agreement. The number of clusters at which the index attains its maximum is a candidate for the optimal number of clusters. The proposed method is applied to simulated bivariate normal data, and further extended for use in circular data. Its performance is compared to the criteria discussed in Tibshirani, Walther, and Hastie (2001). The proposed method is not based on any distributional or data assumption which makes it widely applicable to any type of data that can be clustered using at least two clustering algorithms.

Thyroglossal Duct Cyst—More Than Just an Embryological Remnant

Thyroglossal duct cyst is a congenital malformation that occurs due to incomplete closure of the thyroglossal duct. Apart from being a quiescent embryological remnant, it presents itself clinically at any age and often requires surgical excision. Twenty four patients were encountered at Sri Ramachandra Medical College and Research Institute between June 2004 and June 2009 with thyroglossal duct cyst. The purpose of the study was to analyse their presentation, associated complications including malignancy and the approach to their management. Operative notes, histopathology files and medical records were used for the retrospective analysis of the patients with thyroglossal duct cyst. Thyroid profile and Radioisotope scan were performed on all the patients to document the position of the thyroid gland. Further, these patients were subjected to Ultrasonography and Computerised Tomography neck to conclusively map the location of the cyst. Fine Needle Aspiration of the tissue was done to diagnose malignancy. It was noted that there was maximum clustering of about 62.5% in males. About 45% of them belonged to the paediatric age group. The total number of symptomatic patients was about 59%. Unlike the various studies in literature, most of the thyroglossal duct cyst patients at our hospital were symptomatic. The symptoms were pain, dyspnoea, dysphagia, discharge and recurrence. Malignancy was diagnosed in two patients who were ironically asymptomatic. Other complications were looked into and treated. Thyroglossal duct cyst is a cervical anomaly that is usually found as an asymptomatic painless cystic swelling. However, most of our patients had a symptomatic presentation. Association with malignancy was noted. Surgical intervention was the cure for all the patients.

A proof of the convergence theorem of maximum-entropy clustering algorithm

In this paper, we point out that the counterexample constructed by Yu et al. is incorrect by using scientific computing software Sage. This means that the example cannot negate the convergence theorem of maximum entropy clustering algorithm. Furthermore, we construct an example to negate Theorem 1 in Yu’s paper, and we propose Proposition 3 to prove that the limit of the iterative sequence is a local minimum of the objective function while v varies and u remains stable. Finally, we give a theoretical proof of the convergence theorem of maximum entropy clustering algorithm.

Distributed Energy Optimization for Target Tracking in Wireless Sensor Networks

Energy constraint is an important issue in wireless sensor networks. This paper proposes a distributed energy optimization method for target tracking applications. Sensor nodes are clustered by maximum entropy clustering. Then, the sensing field is divided for parallel sensor deployment optimization. For each cluster, the coverage and energy metrics are calculated by grid exclusion algorithm and Dijkstra's algorithm, respectively. Cluster heads perform parallel particle swarm optimization to maximize the coverage metric and minimize the energy metric. Particle filter is improved by combining the radial basis function network, which constructs the process model. Thus, the target position is predicted by the improved particle filter. Dynamic awakening and optimal sensing scheme are then discussed in dynamic energy management mechanism. A group of sensor nodes which are located in the vicinity of the target will be awakened up and have the opportunity to report their data. The selection of sensor node is optimized considering sensing accuracy and energy consumption. Experimental results verify that energy efficiency of wireless sensor network is enhanced by parallel particle swarm optimization, dynamic awakening approach, and sensor node selection.

Pentapeptide overlapping between human immunodeficiency viruses and Home sapiens proteomes is higher than 90%

We used sequence-to-sequence peptide matching to analyze the similarity level of the HIV-1, HIV-2 and HTLV polyprotein sequences to the human proteome. The following results were obtained: 1) pentapeptides from the viral polyproteins are widely, repeatedly and intensively represented in a large number of human proteins; 2) high level of similarity of viral versus human proteomes is still present even using esa- or eptapeptide motifs as probes for sequence identity scanning; 3) a relatively limited number of viral pentameric fragments (about 10% or less) are unique to the viruses with no similarity to the human host; 4) the distribution of the zero similarity pentapeptides is not random in the three main immunodeficiency virus strains, with a maximum clustering in the HIV-2 p160 aa 1007-1019 sequence. The data are important in relation to the role of molecular mimicry in the genesis and diagnosis of immunodeficiency diseases, and the use of low similarity amino acid sequences as a peptidome platform for anti-HIV vaccine constructs exempt from harmful collateral cross-reactions.