Sparse Cluster Research Articles

Structured multi-view k-means clustering

K-means is a very efficient clustering method and many multi-view k-means clustering methods have been proposed for multi-view clustering during the past decade. However, since k-means have trouble uncovering clusters of varying sizes and densities, these methods suffer from the same performance issues as k-means. Improving the clustering performance of multi-view k-means has become a challenging problem. In this paper, we propose a new multi-view k-means clustering method that is able to uncover clusters in arbitrary sizes and densities. The new method simultaneously performs three tasks, i.e., sparse connection probability matrices learning, prototypes aligning, and cluster structure learning. We evaluate the proposed new method by 5 benchmark datasets and compare it with 11 multi-view clustering methods. The experimental results on both synthetic and real-world experiments show the superiority of our proposed method.

Hair cluster detection model based on dermoscopic images

Introduction: Hair loss has always bothered many people, with numerous individuals potentially facing the issue of sparse hair.Methods: Due to a scarcity of accurate research on detecting sparse hair, this paper proposes a sparse hair cluster detection model based on improved object detection neural network and medical images of sparse hair under dermatoscope to optimize the evaluation of treatment outcomes for hair loss patients. A new Multi-Level Feature Fusion Module is designed to extract and fuse features at different levels. Additionally, a new Channel-Space Dual Attention Module is proposed to consider both channel and spatial dimensions simultaneously, thereby further enhancing the model’s representational capacity and the precision of sparse hair cluster detection.Results: After testing on self-annotated data, the proposed method is proven capable of accurately identifying and counting sparse hair clusters, surpassing existing methods in terms of accuracy and efficiency.Discussion: Therefore, it can work as an effective tool for early detection and treatment of sparse hair, and offer greater convenience for medical professionals in diagnosis and treatment.

An Exploration of the Reform of English Informatisation Teaching in Colleges and Universities Based on Deep Learning Model and Microteaching Mode

Abstract In this paper, we use the cross-layer connectivity of residual networks in deep learning to convert convolutional and fully connected layers into sparse connections and cluster sparse matrices into relatively dense subspaces. Extracted features are used to perform target class prediction and regression of target coordinates using a target detection algorithm to meet the demand for real-time target detection. The model's use resulted in a head-up rate of 83.57% in the classroom, with the least serious students at 0.8 and above. Deep learning technology can enhance students' learning experience in English classrooms by providing personalized learning and a deep learning environment.

Heavy grazing causes plant cluster fragmentation of sparse grasses.

Cleistogenes songorica, as a clustered grass, is the main grassland flora of the Stipa breviflora desert grassland. Some studies have shown that the constructive species of S. breviflora (sparse cluster type) is prone to cluster fragmentation; however, research on C. songorica is relatively rare. Then will the C. songorica plant population (dense cluster type) also have cluster fragmentation under the influence of intense grazing? To answer this question, we used variance analysis and geo-statistical methods. The spatial distribution of C. songorica in S. breviflora desert steppe in Inner Mongolia was measured under four grazing intensities (no grazing, CK, 0 sheep·ha-1·half year-1; light grazing, LG, 0.93 sheep·ha-1·half year-1; moderate grazing, MG, 1.82 sheep·ha-1·half year-1; and heavy grazing, HG, 2.71 sheep·ha-1·half year-1) and four scales (10 cm × 10 cm, 20 cm × 20 cm, 25 cm × 25 cm, 50 cm × 50 cm). We then analyzed C. songorica whether fragmentation was present. The results showed that increased grazing intensity is associated with increased density and decreased height, coverage, and standing crop of C. songorica. The spatial distribution of C. songorica was affected by structural factors, and spatial heterogeneity decreased with increased spatial scale. With increased grazing intensity and spatial scale, the patch area of C. songorica gradually increased and tended toward band distribution. In summary, increased grazing intensity and spatial scale led to weakened heterogeneity of C. songorica spatial distribution and increased consistency.

Efficient Layer-Wise N:M Sparse CNN Accelerator with Flexible SPEC: Sparse Processing Element Clusters

Recently, the layer-wise N:M fine-grained sparse neural network algorithm (i.e., every M-weights contains N non-zero values) has attracted tremendous attention, as it can effectively reduce the computational complexity with negligible accuracy loss. However, the speed-up potential of this algorithm will not be fully exploited if the right hardware support is lacking. In this work, we design an efficient accelerator for the N:M sparse convolutional neural networks (CNNs) with layer-wise sparse patterns. First, we analyze the performances of different processing element (PE) structures and extensions to construct the flexible PE architecture. Second, the variable sparse convolutional dimensions and sparse ratios are involved in the hardware design. With a sparse PE cluster (SPEC) design, the hardware can efficiently accelerate CNNs with the layer-wise N:M pattern. Finally, we employ the proposed SPEC into the CNN accelerator with flexible network-on-chip and specially designed dataflow. We implement hardware accelerators on Xilinx ZCU102 FPGA and Xilinx VCU118 FPGA and evaluate them with classical CNNs such as Alexnet, VGG-16, and ResNet-50. Compared with existing accelerators designed for structured and unstructured pruned networks, our design achieves the best performance in terms of power efficiency.

Load Balancing Control Algorithm of Internet of Things Link Based on Non-Parametric Regression Model

In order to solve the problems of poor channel balance control ability and unable to effectively reduce the output bit error rate in the traditional Internet of things link load balance control methods, a new Internet of things (IoT) link load balance control algorithm based on non-parametric regression model is proposed in this paper. The transmission model of IoT link channel is constructed, and the sparse random cluster analysis method is used to extract the load characteristics of IoT link. According to the load feature extraction results, through the estimated regression function of known data features, a non-parametric regression model is constructed, and the fuzzy cyclic iterative control is used to realize the load balancing control of the Internet of things link. The experimental results show that this method has strong channel balance control ability, low output bit error rate, the maximum average link utilisation can reach 1, and the maximum output bit error rate is only 102, which improves the stability of the Internet of things.

Density peaks clustering algorithm based on fuzzy and weighted shared neighbor for uneven density datasets

Uneven density data refers to data with a certain difference in sample density between clusters. The local density of density peaks clustering algorithm (DPC) does not consider the effect of sample density difference between clusters of uneven density data, which may lead to wrong selection of cluster centers; the algorithm allocation strategy makes it easy to incorrectly allocate samples originally belonging to sparse clusters to dense clusters, which reduces clustering efficiency. In this study, we proposed the density peaks clustering algorithm based on fuzzy and weighted shared neighbor for uneven density datasets (DPC-FWSN). First, a nearest neighbor fuzzy kernel function is obtained by combining K-nearest neighbor and fuzzy neighborhood. Then, local density is redefined by the nearest neighbor fuzzy kernel function. The local density can better characterize the distribution characteristics of the sample by balancing the contribution of sample density in dense and sparse areas, in order to avoid the situation that the sparse cluster does not have a cluster center. Finally, the allocation strategy for weighted shared neighbor similarity is proposed to optimize the sample allocation at the boundary of the sparse cluster. Experiments are performed on IDPC-FA, FKNN-DPC, FNDPC, DPCSA and DPC for uneven density datasets, complex morphologies datasets and real datasets. The clustering results demonstrate that DPC-FWSN effectively handles datasets with uneven density distribution.

A Novel Autoencoder Deep Architecture for Detecting the Outlier in Heterogeneous Data Sets

Neighborhood-based unsupervised approaches like LDOF, LOF and symmetric neighborhood (INFLO) have proven effective over decades. These techniques principally utilize the information of either the [Formula: see text]-nearest neighbor or the reverse [Formula: see text]-nearest neighbors to detect the outlierness of each object in a data set. However, these methodologies fail to detect genuine outliers in heterogeneous data sets located between two dense clusters, between dense and sparse clusters or the scattered data sets. In addition, LOF treats a normal point of a sparse cluster as an outlier if the sparse cluster is close to a dense cluster. This paper proposes a novel autoencoder deep learning architecture to overcome the limitations of the aforementioned techniques. In the proposed approach, we identify the potential outliers intelligently from a given data set and mark them to generate training samples for the autoencoder. These marked points are not included in the training samples. Finally, the trained autoencoder is used to compute the outlierness of each data point in the whole data set (training samples + marked points). Experimental results with synthetic and real-world data sets show that the proposed model outperforms these widely applied techniques along with state-of-the-art works.

Motor patterns of the impaired upper limb in children with unilateral cerebral palsy performing bimanual tasks

BackgroundUpper limb movement patterns have not yet been identified in bimanual conditions despite the difficulties children with unilateral cerebral palsy have performing bimanual activities. The aim was to identify specific motor patterns from kinematic deviations during bimanual tasks in this population. MethodsTwenty children with unilateral cerebral palsy and 20 age-matched, typically developing children performed the five tasks of a 3D bimanual protocol. To evaluate upper limb kinematic deviations, 10 Arm Variable Scores were calculated for the affected /non-dominant upper limb of each participant for each task. Sparse K-means cluster analysis was applied to the 50 Arm Variable Scores of all the children to identify motor patterns and determining variables. Clinical tests of impairment (muscle strength, selectivity, spasticity) and function (Assisting hand assessment, Abilhand-Kids) were compared between the clusters obtained. FindingsThree different motor patterns were identified using the data from all the children: mild, proximal-distal and proximal-distal with trunk. The most important cluster determinants were the Arm Variable Scores for pronation-supination and wrist extension. In the cerebral palsy group, scores of impairments (p < .01) and function (Assisting Hand Assessment [p < .001] and Abilhand-Kids [p = .004]) differed for each motor pattern. Supination and wrist extension deviations differed significantly between the groups (p < .001). InterpretationDuring performance of bimanual tasks, children with unilateral cerebral palsy used distinct motor patterns that each corresponded to a specific clinical profile. Elbow-wrist deviations were the largest and most decisive and were specific to the cerebral palsy group: they should be the target of interventions to enhance bimanual function.ClinicalTrials.gov identifier: NCT03888443

Clustering of Short Texts Based on Dynamic Adjustment for Contrastive Learning

Faced with the large amount of unlabeled short text data appearing on the Internet, it is necessary to categorize them using clustering that can divide text into several clusters based on similarity degree of text semantics. Recently, combining clustering with contrastive learning has been the focus of clustering research. Due to the excellent representation learning ability of contrastive learning, clustering achieves better results on short texts that are high-dimensional and sparse. However, contrastive learning pays more attention to general feature representation at the instance-level and ignores the semantic-level correlation of data belonging to same cluster in clustering. The inconsistent training objectives of contrastive learning and clustering lead to lower confidence of clustering results and sparse cluster space. To improve this problem, we propose a clustering method based on Dynamic Adjustment for Contrastive Learning (DACL). The method smoothly transitions loss weight of model from contrastive learning to clustering during training and filters negative samples in contrastive learning by the pseudo-labels generated by clustering. To demonstrate the effectiveness of the method, DACL is compared with eight short text clustering models on eight datasets. The results show that we achieve considerable performance improvements on most datasets compared to state-of-the-art short text clustering methods. In addition, The effectiveness of loss smooth transition and negative filtering is proved by ablation experiments.

Tracing the magnetic field morphology of the LDN 1172/1174 cloud complex

Context. The LDN 1172/1174 cloud complex in the Cepheus Flare region presents a hub-filament structure with the reflection nebula, NGC 7023, illuminated by a Herbig Be star, HD 200775, which consists of the hub with a ~5 pc long narrow filament attached to it. Formation of a sparse cluster of low- and intermediate-mass stars is presently taking place in the hub. Aims. The aim of this work is to map the magnetic field geometry of LDN 1172/1174 to understand the role played by the field lines in the formation of the molecular cloud. Methods. Unpolarized background stars can be used to measure dichroically polarized light produced by the magnetically aligned grains present in molecular clouds. As these dust grains get aligned with their minor axes parallel to the ambient magnetic field, the polarization measurements can provide the plane-of-sky component of the magnetic field. We made R-band polarization measurements of 249 stars projected on the entire LDN 1172/1174 cloud complex to map the geometry of the magnetic field of this region. Results. The magnetic field geometry constructed from our R-band polarization measurements is found to be parallel to the elongated structure inferred from the column density distribution of the cloud produced using the Herschel images. Our R-band polarization measurements are found to be in good agreement with those obtained from Planck. There is evidence of a possible distortion of the magnetic fields toward the northwestern part of the cloud by HD 200775. The magnetic field strength is estimated as ~30 μG. The estimated star formation rate (SFR)/mass of 2.0 ± 1.3% Myr−1 and 0.4 ± 0.3% Myr−1 for LDN 1172/1174 and the neighboring cloud complex, LDN 1147/1158, respectively, are found to be consistent with the mean SFR/mass found for the clouds with magnetic field orientations parallel and perpendicular to their elongated structures, respectively. These results support earlier findings that the clouds with magnetic field lines parallel to their long axes seem to have higher SFRs compared to those with the magnetic field orientation perpendicular to the cloud elongation.

462 Redefining Positive Airway Pressure Adherence Phenotypes Utilizing Deep Neural Networks and Unsupervised Clustering

Abstract Introduction Improving positive airway pressure (PAP) adherence is crucial to obstructive sleep apnea (OSA) treatment success. We have previously shown the potential of utilizing Deep Neural Network (DNN) models to accurately predict future PAP usage, based on predefined compliance phenotypes, to enable early patient outreach and interventions. These phenotypes were limited, based solely on usage patterns. We propose an unsupervised learning methodology for redefining these adherence phenotypes in order to assist with the creation of more precise and personalized patient categorization. Methods We trained a DNN model to predict PAP compliance based on daily usage patterns, where compliance was defined as the requirement for 4 hours of PAP usage a night on over 70% of the recorded nights. The DNN model was trained on N=14,000 patients with 455 days of daily PAP usage data. The latent dimension of the trained DNN model was used as a feature vector containing rich usage pattern information content associated with overall PAP compliance. Along with the 455 days of daily PAP usage data, our dataset included additional patient demographics such as age, sex, apnea-hypopnea index, and BMI. These parameters, along with the extracted usage patterns, were applied together as inputs to an unsupervised clustering algorithm. The clusters that emerged from the algorithm were then used as indicators for new PAP compliance phenotypes. Results Two main clusters emerged: highly compliant and highly non-compliant. Furthermore, in the transition between the two main clusters, a sparse cluster of struggling patients emerged. This method allows for the continuous monitoring of patients as they transition from one cluster to the other. Conclusion In this research, we have shown that by utilizing historical PAP usage patterns along with additional patient information we can identify PAP specific adherence phenotypes. Clinically, this allows focus of PAP adherence program resources to be targeted early on to patients susceptible to treatment non-adherence. Furthermore, the transition between the two main phenotypes can also indicate when personalized intervention is necessary to maximize treatment success and outcomes. Lastly, providers can transition patients in the highly non-compliant group more quickly to alternative therapies. Support (if any):

Improved Density Peaks Clustering Based on Natural Neighbor Expanded Group

Density peaks clustering (DPC) is an advanced clustering technique due to its multiple advantages of efficiently determining cluster centers, fewer arguments, no iterations, no border noise, etc. However, it does suffer from the following defects: (1) difficult to determine a suitable value of its crucial cutoff distance parameter, (2) the local density metric is too simple to find out the proper center(s) of the sparse cluster(s), and (3) it is not robust that parts of prominent density peaks are remotely assigned. This paper proposes improved density peaks clustering based on natural neighbor expanded group (DPC-NNEG). The cores of the proposed algorithm contain two parts: (1) define natural neighbor expanded (NNE) and natural neighbor expanded group (NNEG) and (2) divide all NNEGs into a goal number of sets as the final clustering result, according to the closeness degree of NNEGs. At the same time, the paper provides the measurement of the closeness degree. We compared the state of the art with our proposal in public datasets, including several complex and real datasets. Experiments show the effectiveness and robustness of the proposed algorithm.

GDPC: generalized density peaks clustering algorithm based on order similarity

Clustering is a fundamental approach to discover the valuable information in data mining and machine learning. Density peaks clustering is a typical density based clustering and has received increasing attention in recent years. However DPC and most of its improvements still suffer from some drawbacks. For example, it is difficult to find peaks in the sparse cluster regions; assignment for the remaining points tends to cause Domino effect, especially for complicated data. To address the above two problems, we propose generalized density peaks clustering algorithm (GDPC) based on a new order similarity, which is calculated by the order rank of Euclidean distance between two samples. The order similarity can help us to find peaks in the sparse regions. In addition, a two-step assignment is used to weaken Domino effect. In general, GDPC can not only discover clusters in datasets regardless of different sizes, dimensions and shapes, but also address the above two issues. Several experiments on datasets, including Lung, COIL20, ORL, USPS, Mnist, breast and Vote, show that our algorithm is effective in most cases.

A new look at Sco OB1 association with Gaia DR2

ABSTRACT We present and discuss photometric optical data in the area of the OB association Sco OB1 covering about 1 deg2. UBVI photometry is employed in tandem with Gaia DR2 data to investigate the three-dimensional structure and the star formation history of the region. By combining parallaxes and proper motions, we identify seven physical groups located between the young open cluster NGC 6231 and the bright nebula IC 4628. The most prominent group coincides with the sparse open cluster Trumpler 24. We confirm the presence of the intermediate-age star cluster VdB-Hagen 202, which is unexpected in this environment, and provide for the first time estimates of its fundamental parameters. After assessing individual groups membership, we derive mean proper motion components, distances, and ages. The seven groups belong to two different families. To the younger family (family I) belong several pre-main-sequence (PMS) stars as well. These are evenly spread across the field, and also in front of VdB-Hagen 202. VdB-Hagen 202, and two smaller, slightly detached, groups of similar properties form family II, which do not belong to the association, but are caught in the act of passing through it. As for the younger population, this forms an arc-like structure from the bright nebula IC 4628 down to NGC 6231, as previously found. Moreover, the PMS stars density seems to increase from NGC 6231 northward to Trumpler 24.

A Domain Adaptive Density Clustering Algorithm for Data With Varying Density Distribution

As one type of efficient unsupervised learning methods, clustering algorithms have been widely used in data mining and knowledge discovery with noticeable advantages. However, clustering algorithms based on density peak have limited clustering effect on data with varying density distribution (VDD), equilibrium distribution (ED), and multiple domain-density maximums (MDDM), leading to the problems of sparse cluster loss and cluster fragmentation. To address these problems, we propose a Domain-Adaptive Density Clustering (DADC) algorithm, which consists of three steps: domain-adaptive density measurement, cluster center self-identification, and cluster self-ensemble. For data with VDD features, clusters in sparse regions are often neglected by using uniform density peak thresholds, which results in the loss of sparse clusters. We define a domain-adaptive density measurement method based on K-Nearest Neighbors (KNN) to adaptively detect the density peaks of different density regions. We treat each data point and its KNN neighborhood as a subgroup to better reflect its density distribution in a domain view. In addition, for data with ED or MDDM features, a large number of density peaks with similar values can be identified, which results in cluster fragmentation. We propose a cluster center self-identification and cluster self-ensemble method to automatically extract the initial cluster centers and merge the fragmented clusters. Experimental results demonstrate that compared with other comparative algorithms, the proposed DADC algorithm can obtain more reasonable clustering results on data with VDD, ED and MDDM features. Benefitting from a few parameter requirements and non-iterative nature, DADC achieves low computational complexity and is suitable for large-scale data clustering.

ASAP: Adaptive Structure Aware Pooling for Learning Hierarchical Graph Representations

Graph Neural Networks (GNN) have been shown to work effectively for modeling graph structured data to solve tasks such as node classification, link prediction and graph classification. There has been some recent progress in defining the notion of pooling in graphs whereby the model tries to generate a graph level representation by downsampling and summarizing the information present in the nodes. Existing pooling methods either fail to effectively capture the graph substructure or do not easily scale to large graphs. In this work, we propose ASAP (Adaptive Structure Aware Pooling), a sparse and differentiable pooling method that addresses the limitations of previous graph pooling architectures. ASAP utilizes a novel self-attention network along with a modified GNN formulation to capture the importance of each node in a given graph. It also learns a sparse soft cluster assignment for nodes at each layer to effectively pool the subgraphs to form the pooled graph. Through extensive experiments on multiple datasets and theoretical analysis, we motivate our choice of the components used in ASAP. Our experimental results show that combining existing GNN architectures with ASAP leads to state-of-the-art results on multiple graph classification benchmarks. ASAP has an average improvement of 4%, compared to current sparse hierarchical state-of-the-art method. We make the source code of ASAP available to encourage reproducible research 1.

Stellar population astrophysics (SPA) with the TNG. Characterization of the young open cluster ASCC 123

Star clusters are key to understand the stellar and Galactic evolution. ASCC 123 is a little-studied, nearby and very sparse open cluster. We performed the first high-resolution spectroscopic study of this cluster in the framework of the SPA (Stellar Population Astrophysics) project with GIARPS at the TNG. We observed 17 stars, five of which turned out to be double-lined binaries. Three of the investigated sources were rejected as members on the basis of astrometry and lithium content. For the remaining single stars we derived the stellar parameters, extinction, radial and projected rotational velocities, and chemical abundances for 21 species with atomic number up to 40. From the analysis of single main-sequence stars we found an average extinction $A_V\simeq 0.13$ mag and a median radial velocity of about $-5.6$ km/s. The average metallicity we found for ASCC 123 is [Fe/H]$\simeq+0.14\pm 0.04$, which is in line with that expected for its Galactocentric distance. The chemical composition is compatible with the Galactic trends in the solar neighborhood within the errors. From the lithium abundance and chromospheric H$\alpha$ emission we found an age similar to that of the Pleiades, which agrees with that inferred from the Hertzsprung-Russell and color-magnitude diagrams.

Piecewise Flat Embedding for Image Segmentation.

We introduce a new multi-dimensional nonlinear embedding-Piecewise Flat Embedding (PFE)-for image segmentation. Based on the theory of sparse signal recovery, piecewise flat embedding with diverse channels attempts to recover a piecewise constant image representation with sparse region boundaries and sparse cluster value scattering. The resultant piecewise flat embedding exhibits interesting properties such as suppressing slowly varying signals, and offers an image representation with higher region identifiability which is desirable for image segmentation or high-level semantic analysis tasks. We formulate our embedding as a variant of the Laplacian Eigenmap embedding with an $L_{1,p} (0<p\leq 1)$L1,p(0<p≤1) regularization term to promote sparse solutions. First, we devise a two-stage numerical algorithm based on Bregman iterations to compute $L_{1,1}$L1,1-regularized piecewise flat embeddings. We further generalize this algorithm through iterative reweighting to solve the general $L_{1,p}$L1,p-regularized problem. To demonstrate its efficacy, we integrate PFE into two existing image segmentation frameworks, segmentation based on clustering and hierarchical segmentation based on contour detection. Experiments on four major benchmark datasets, BSDS500, MSRC, Stanford Background Dataset, and PASCAL Context, show that segmentation algorithms incorporating our embedding achieve significantly improved results.

Clumsy encounters or graceful mergers?

Observations of two sequences of blue stragglers in a young, sparse star cluster in the Large Magellanic Cloud re-opens the debate about the dominant formation mechanism of these anomalous stars.