Unsupervised Pattern Classification Research Articles

Sustainable vertically-oriented graphene-electrode memristors for neuromorphic applications

Neuromorphic computing, an innovative field in electronic and computing engineering, aims to enhance computing paradigms by simulating brain processes. Memristors, a two-terminal device, hold promise in revolutionising neuromorphic architectures by circumventing the Von-Neumann bottleneck. The performance and applicability of memristors heavily rely on the materials and fabrication processes employed. Graphene exhibits unique properties that can be leveraged in memristor design. Moreover, graphene stands out as a material with the potential for large-scale, sustainable production through Plasma Enhanced Chemical Vapour Deposition (PECVD). Notably, the properties of graphene-electrode memristors vary with minor structural differences induced by different PECVD temperatures. This paper reports the synthesis of graphene electrodes by time- and cost-effective PECVD from a sustainable plant extract for memristors. In addition, this paper delves into investigating how these structural variations impact the properties of graphene memristors and explores their potential exploitation in neuromorphic applications for implementing the well-known Spike Timing Dependent Plasticity (STDP) learning mechanism. The paper also utilises the developed STDP learning to perform an unsupervised spike-based pattern classification task.

Unsupervised Classification of Spike Patterns with the Loihi Neuromorphic Processor

Unsupervised Classification of Spike Patterns with the Loihi Neuromorphic Processor

Topological data analysis of spatial patterning in heterogeneous cell populations: clustering and sorting with varying cell-cell adhesion

Different cell types aggregate and sort into hierarchical architectures during the formation of animal tissues. The resulting spatial organization depends (in part) on the strength of adhesion of one cell type to itself relative to other cell types. However, automated and unsupervised classification of these multicellular spatial patterns remains challenging, particularly given their structural diversity and biological variability. Recent developments based on topological data analysis are intriguing to reveal similarities in tissue architecture, but these methods remain computationally expensive. In this article, we show that multicellular patterns organized from two interacting cell types can be efficiently represented through persistence images. Our optimized combination of dimensionality reduction via autoencoders, combined with hierarchical clustering, achieved high classification accuracy for simulations with constant cell numbers. We further demonstrate that persistence images can be normalized to improve classification for simulations with varying cell numbers due to proliferation. Finally, we systematically consider the importance of incorporating different topological features as well as information about each cell type to improve classification accuracy. We envision that topological machine learning based on persistence images will enable versatile and robust classification of complex tissue architectures that occur in development and disease.

Real-Time Monitoring of Miniaturized Thermal Food Processing by Advanced Mass Spectrometric Techniques

Mass spectrometry is a popular and powerful analytical tool to study the effects of food processing. Industrial sampling, real-life sampling, or challenging academic research on process-related volatile and aerosol research often demand flexible, time-sensitive data acquisition by state-of-the-art mass analyzers. Here, we show a laboratory-scaled, miniaturized, and highly controllable setup for the online monitoring of aerosols and volatiles from thermal food processing based on dielectric barrier discharge ionization (DBDI) mass spectrometry (MS). We demonstrate the opportunities offered by the setup from a foodomics perspective to study emissions from the thermal processing of wheat bread rolls at 210 °C by Fourier transformation ion cyclotron resonance MS. As DBDI is an emerging technology, we compared its ionization selectivity to established atmospheric pressure ionization tools: we found DBDI preferably ionizes saturated, nitrogenous compounds. We likewise identified a sustainable overlap in the selectivity of detected analytes with APCI and electrospray ionization (ESI). Further, we dynamically recorded chemical fingerprints throughout the thermal process. Unsupervised classification of temporal response patterns was used to describe the dynamic nature of the reaction system. Compared to established tools for real-time MS, our setup permits one to monitor chemical changes during thermal food processing at ultrahigh resolution, establishing an advanced perspective for real-time mass spectrometric analysis of food processing.

Retracted: Real-Time Water and Electricity Consumption Monitoring Using Machine Learning Techniques

This work studies the automatic classification of water consumption patterns and electrical devices, both supervised and unsupervised. This involves training machine learning algorithms to identify normal and abnormal water consumption patterns and differentiate between different types of electrical devices. We performed an unsupervised classification of consumer water patterns in direct and indirect ways. The first is to use the raw consumption patterns obtained directly from the server. The second one corresponds to the use of the sampled consumption patterns. This classification is performed using hierarchical bottom-up classification and a self-organizing map. A probabilistic analysis of daily water consumption is performed to extract the percentage of daily consumption with the most information. It enables us to identify water consumption patterns more quickly by reducing the number of data points for each daily pattern, allowing us to recognize and classify anomalous behaviors as soon as feasible. Then, the signatures of electrical devices are classified using three ML algorithms: multilayer perceptron (MLP), k nearest neighbor (KNN), and decision tree (DT). Furthermore, assembly approaches are also studied. These are based on the OAA (One Against All) principle, which presents one class against all other classes, and the ECOC (Error-Correcting Output Codes) philosophy, which allows the classification error to be corrected. According to the bias/variance trade-off, both techniques enhanced classification by ensuring accuracy and generalization. A more in-depth analysis of the properties of the electrical devices is handled with an esemplastic approach based on gradient-boosting decision trees. The features of electrical devices can be extracted using this analysis based on the nature of their harmonic signatures

DPCF: A framework for imputing missing values and clustering data in drug discovery process

The advent of modern Internet of Things (IoT) architectures has led to ease in data collection and availability. The data generated from such architectures are of large volume and dimensionality. As a result, data missingness and data labeling are the commonly occurring problems in the data collection process when data volume is too large. Data clustering is a commonly used unsupervised pattern classification technique that helps in identifying the hidden structure of the datasets and clusters or groups similar data items together. In context to chemometrics, clustering techniques play a significant role in identifying the structure-property relationships and structure-activity relationships among different compounds in drug discovery process. The quantitative structure-property relationships (QSPR) are based on the similar property principle which states that if the compounds are clustered together based on structural descriptors, the compounds within the same cluster will have similar properties. The quantitative structure-activity relationships (QSAR) help in determining the empirical relationships between chemical structure and biological activity of a set of identical compounds. Hence, a variety of compounds can be divided into homogeneous subsets using cluster analysis. The effectiveness of any clustering algorithm mostly depends on how efficiently the initial cluster centers are identified. Traditional techniques that mostly rely on the random selection of initial cluster centers and different parameter settings may result in distinct clusters for the same dataset. Moreover, if the initial cluster centers selected are outliers then the clusters formed are of poor quality. The quality of each cluster can be determined by using distinct cluster validation indices (CVIs). This paper aims to solve the problems of data missingness as well as data labeling for unsupervised datasets. In continuation to the previous study where the NMVI (Nullify the Missing Values before Imputation) technique efficiently imputes the missing values in different datasets, in this paper the proposed DPCF (Data Partitioning-based Clustering Framework) framework utilize the NMVI technique to impute the missing values and a novel Z-Clust clustering algorithm is proposed that efficiently clusters the unlabeled data samples. The integration of the NMVI technique and Z-Clust clustering technique in the proposed framework makes it well suited for the analysis of unsupervised datasets having missing values. The performance evaluation of the proposed Z-Clust clustering technique is done by using five standard CVIs and the results are compared with the existing clustering techniques. The experimental results depict that the proposed Z-Clust clustering technique shows better cluster formation as compared to the existing clustering techniques. Henceforth, the proposed DPCF framework is well suited for the analysis of datasets without labels and having missing values.

Unsupervised novelty pattern classification of shmoo plots for visualizing the test results of integrated circuits

Shmoo plots are visual tools for verifying the performance of integrated circuit devices, where each cell in a plot records whether the examined device operates normally under the test condition. When identifying the device state, the overall pass/fail pattern appearing in the shmoo plot is more important than the test results for individual conditions. Because similar shmoo plots indicate similar device characteristics, defect causes, and process peculiarities, engineers can analyze device quality and defect causes by classifying shmoo plot patterns. Most mass-produced devices have high and stable yields, whereas defect devices are incredibly scarce. If engineers classify numerous device plots manually at a semiconductor test site, significant time and resources will be required, and the result will likely vary based on the engineers’ experience. Therefore, shmoo plot usage is limited unless an automatic classification model is adopted. Moreover, training high-performance pattern classifiers that do not overfit the models is difficult because shmoo plots contain high-dimensional data and unlabeled, multiclass imbalanced datasets, where the number of defects is smaller than that of normal plots and pattern labels are seldom assigned. In this study, we propose a novel feature extraction process and a two-stage clustering process to distinguish novel shmoo plot patterns. Actual shmoo plots obtained from a wafer test stage are used to compare the experimental results obtained via the proposed method and conventional methods, and they verify the superiority of the proposed method.

A new validity clustering index-based on finding new centroid positions using the mean of clustered data to determine the optimum number of clusters

Clustering, an unsupervised pattern classification method, plays an important role in identifying input dataset structures. It partitions input datasets into clusters or groups where either the optimum number of clusters is known in prior or automatically determined. In the case of automatic clustering, the performance is evaluated using a cluster validity index (CVI), which determines the optimum number of clusters in the data. From previous works, the improper cluster centroids positioning produced by clustering algorithms could reduce the performance of the validation process and performance produced by the previous state-of-the-art CVIs. In addition, those previous CVIs can only work properly with certain clustering algorithms and simple datasets structures, which their performances will reduce if they are applied to other clustering algorithms as well as more complex datasets. This study proposes an efficient CVI, namely, the validity clustering index based on finding the mean of clustered data (VCIM). The proposed approach combines the properties of the score function index and the mean to determine new cluster centroid positions. The performance of the VCIM index is compared with well-known CVIs on both artificial and real-life datasets. The obtained results on artificial datasets show that the proposed VCIM index outperforms the other CVIs in determining the true number of clusters for the five conventional clustering algorithms, namely, K-means, Fuzzy C-mean, agglomerative hierarchical average linkage clustering, variance-based differential evolution, and density peaks clustering and Particle swarm optimization (PDPC) algorithms. For the 14 real-word datasets, the proposed VCIM index correctly determined the optimum number of clusters for 11 out of 14 for the K-means clustering algorithm, 9 out of 14 for both Fuzzy clustering and agglomerative hierarchical average linkage clustering algorithms, 12 out of 14 for the variance-based differential evolution algorithm and 11 out of 14 datasets for PDPC. The obtained results using the proposed VCIM show its significance when combined with clustering algorithms and nominate its potential in various clustering applications.

Nanoscale mosaicity revealed in peptide microcrystals by scanning electron nanodiffraction

Changes in lattice structure across sub-regions of protein crystals are challenging to assess when relying on whole crystal measurements. Because of this difficulty, macromolecular structure determination from protein micro and nanocrystals requires assumptions of bulk crystallinity and domain block substructure. Here we map lattice structure across micron size areas of cryogenically preserved three−dimensional peptide crystals using a nano-focused electron beam. This approach produces diffraction from as few as 1500 molecules in a crystal, is sensitive to crystal thickness and three−dimensional lattice orientation. Real-space maps reconstructed from unsupervised classification of diffraction patterns across a crystal reveal regions of crystal order/disorder and three−dimensional lattice tilts on the sub-100nm scale. The nanoscale lattice reorientation observed in the micron-sized peptide crystal lattices studied here provides a direct view of their plasticity. Knowledge of these features facilitates an improved understanding of peptide assemblies that could aid in the determination of structures from nano- and microcrystals by single or serial crystal electron diffraction.

Clustering Analysis of Normal Strength Concretes Produced with Different Aggregate Types

Abstract Concrete is the most commonly used structural material, which is composed of individual base materials. The compressive strength of concrete is important to understand for activities like construction arrangement, prestressing operations, proportioning new mixtures and quality assurance. Concrete has a problem known as Clustering, which is the unsupervised classification of patterns into clusters. The clustering problem has been addressed by several researchers in many contexts and various disciplines; this shows that clustering uses many areas and is an important step in data analysis. In this study, concrete samples with different aggregate types and normal strength were produced. Clustering analysis was performed on the effect of aggregate species for its mechanical strength. According to the result of clustering analysis, it was found that there were three different groups, the study predicts that clustering is not related to the origins of the rocks but is clustered between the strength of the rocks and the mechanical strengths of the concretes produced within these rocks. The common feature of L, RL and SG aggregates in the first cluster is that it is a sedimentary rock. The aggregate of the concrete in the second cluster is that it is a volcanic superficial rock. The common feature of DO and TB rocks in the third cluster is that it has high density. As a result, it was determined that the clay made in normal class concrete is related to the mechanical strength of rocks.

An effective partitional clustering algorithm based on new clustering validity index

As an unsupervised pattern classification method, clustering partitions the input datasets into groups or clusters. It plays an important role in identifying the natural structure of the target datasets. Now, it has been widely used in data mining, pattern recognition, image processing and so on. However, due to different settings of the parameters and random selection of initial centers, traditional clustering algorithms may produce different clustering partitions for a single dataset. Clustering validity index (CVI) is an important method for evaluating the effect of clustering results generated by clustering algorithms. However, many of the existing CVIs suffer from complex computation, low time efficiency and narrow range of applications. In order to make clustering algorithms more stable, traditional K-means is firstly improved by the density parameters based initial center selection method other than randomly selecting initial centers. Then, in order to enlarge the application range of clustering and better evaluate the clustering partition results, a new variance based clustering validity index (VCVI) from the point of view of spatial distribution of datasets is designed. Finally, a new partitional clustering algorithm integrated with the improved K-means algorithm and the newly introduced VCVI is designed to optimize and determine the optimal clustering number (Kopt) for a wide range of datasets. Furthermore, the commonly used empirical rule Kmax⩽n is reasonably explained by the newly designed VCVI. The new algorithm integrated with VCVI is compared with traditional algorithms integrated with five commonly used CVIs. The experimental results show that our new clustering method is more accurate and stable while consuming relatively lower running time.

Developing an Ornamental Fish Warehousing System Based on Big Video Data

We have developed an ornamental fish warehousing (OFWare) system based on big video data. The system is an application paradigm of information and communication technologies for traditional industries, specifically in the fields of aquaculture and agriculture. Live creatures are the main products of these industries, raising challenges for warehouse management. Warehousing of high unit-price ornamental fishes such as koi, stingray, and arowana is even more difficult since, in addition to counting and classification, such warehousing requires the identification of individual animals whose shapes and texture patterns vary as they grow. Therefore, rather than using invasive RFID-based systems, we combine mobile cloud computing and big data analytics techniques including image and video collection and transmission using handheld mobile devices, unsupervised texture pattern classification of fish tank videos, fish image retrieval, and statistical analysis. The proposed system is scalable based on a Hadoop framework and a small set of a single name-nodes and data-nodes can identify a particular fish among 500,000 koi in 7 seconds. The proposed warehousing system can form the basis for the development of breeding histories, anti-forgery certificate, and aquaculture business intelligence.

Genomic signal processing for DNA sequence clustering.

Genomic signal processing (GSP) methods which convert DNA data to numerical values have recently been proposed, which would offer the opportunity of employing existing digital signal processing methods for genomic data. One of the most used methods for exploring data is cluster analysis which refers to the unsupervised classification of patterns in data. In this paper, we propose a novel approach for performing cluster analysis of DNA sequences that is based on the use of GSP methods and the K-means algorithm. We also propose a visualization method that facilitates the easy inspection and analysis of the results and possible hidden behaviors. Our results support the feasibility of employing the proposed method to find and easily visualize interesting features of sets of DNA data.

Coclustering of Multidimensional Big Data: A Useful Tool for Genomic, Financial, and Other Data Analysis

The analysis of a multidimensional data array is necessary in many applications. Although a data set can be very large, it is possible that meaningful and coherent patterns embedded in the data array are much smaller in size. For example, in genomic data, we may want to find a subset of genes that coexpress under a subset of conditions. In this article, I will explain coclustering algorithms for solving the coherent pattern-detection problem. In these methods, a coherent pattern corresponds to a low-rank matrix or tensor and can be represented as an intersection of hyperplanes in a high-dimensional space. We can then extract coherent patterns from the large data array by detecting hyperplanes. Examples will be provided to demonstrate the effectiveness of the coclustering algorithms for solving unsupervised pattern classification problems.

Classification of Trajectories Using Category Maps and U-Matrix to Predict Interests Used for Event Sites

This paper presents a method for clas- sification and recognition of behavior patterns based on interest from human trajectories at an event site. Our method creates models using Hidden Markov Models (HMMs) for each human trajectory quantized using One-Dimensional Self-Organizing Maps (1D- SOMs). Subsequently, we apply Two-Dimensional SOMs (2D-SOMs) for unsupervised classification of behavior patterns from features according to the dis- tance between models. Furthermore, we use a Unified distance Matrix (U-Matrix) for visualizing category boundaries based on the Euclidean distance between weights of 2D-SOMs. Our method extracts typical be- havior patterns and specific behavior patterns based on interest as ascertained using questionnaires. Then our method visualizes relations between these patterns. We evaluated our method based on Cross Validation (CV) using only the trajectories of typical behavior patterns. The recognition accuracy improved 9.6% over that of earlier models. We regard our method as useful to estimate interest from behavior patterns at an event site.

Unsupervised classification of neocortical activity patterns in neonatal and pre-juvenile rodents.

Flexible communication within the brain, which relies on oscillatory activity, is not confined to adult neuronal networks. Experimental evidence has documented the presence of discontinuous patterns of oscillatory activity already during early development. Their highly variable spatial and time-frequency organization has been related to region specificity. However, it might be equally due to the absence of unitary criteria for classifying the early activity patterns, since they have been mainly characterized by visual inspection. Therefore, robust and unbiased methods for categorizing these discontinuous oscillations are needed for increasingly complex data sets from different labs. Here, we introduce an unsupervised detection and classification algorithm for the discontinuous activity patterns of rodents during early development. For this, in a first step time windows with discontinuous oscillations vs. epochs of network “silence” were identified. In a second step, the major features of detected events were identified and processed by principal component analysis for deciding on their contribution to the classification of different oscillatory patterns. Finally, these patterns were categorized using an unsupervised cluster algorithm. The results were validated on manually characterized neonatal spindle bursts (SB), which ubiquitously entrain neocortical areas of rats and mice, and prelimbic nested gamma spindle bursts (NG). Moreover, the algorithm led to satisfactory results for oscillatory events that, due to increased similarity of their features, were more difficult to classify, e.g., during the pre-juvenile developmental period. Based on a linear classification, the optimal number of features to consider increased with the difficulty of detection. This algorithm allows the comparison of neonatal and pre-juvenile oscillatory patterns in their spatial and temporal organization. It might represent a first step for the unbiased elucidation of activity patterns during development.

ARTIFICIAL BEE COLONY ALGORITHM INTEGRATED WITH FUZZY C-MEAN OPERATOR FOR DATA CLUSTERING

Clustering task aims at the unsupervised classifica tion of patterns in different groups. To enhance th e quality of results, the emerging swarm-based algorithms now-a-days become an alternative to the conventional clustering methods. In this study, an optimization method based on the swarm intelligence algorithm is proposed for the purpose of clustering. The signifi cance of the proposed algorithm is that it uses a F uzzy CMeans (FCM) operator in the Artificial Bee Colony ( ABC) algorithm. The area of action of the FCM operator comes at the scout bee phase of the ABC algorithm as the scout bees are introduced by the FCM operator . The experimental results have shown that the proposed a pproach has provided significant results in terms o f the quality of solution. The comparative study of the p roposed approach with existing algorithms in the li terature using the datasets from UCI Machine learning repository is satisfactory.

INVESTIGATION OF COMBINATIONAL CLUSTERING INDICES IN ARTIFICIAL IMMUNE MULTI‐OBJECTIVE CLUSTERING

Clustering validity indices play a core role in the unsupervised pattern classification. To date, some indices have been proposed and their individual performances were compared on different artificial clustering test instances and real image segmentation tasks. However, little focus has been placed on the issue that how about their combinational performances are, although multi‐objective optimization has attracted much interest from current researchers in unsupervised classification. Here, we firstly evaluate the performance of five state‐of‐the‐art clustering indices with different characteristics in a single‐objective optimization algorithm designed in artificial immune system (AIS). Then, a multi‐objective optimization algorithm in AIS with fair ability of adaptability and diversity maintaining is introduced in this study. After that, the clustering performances of combinational clustering indices are investigated in the multi‐objective optimization framework. To test the effectiveness of the combinational clustering validity indices, 27 benchmark functions with different geometric structures and three complicated remote sensing images are employed in this study. Based on the computer simulations, some meaningful empirical guidelines are obtained for selecting the suitable combinational clustering indices for formulating an effective and robust multi‐objective clustering algorithm in different clustering tasks.

A survey: hybrid evolutionary algorithms for cluster analysis

Clustering is a popular data analysis and data mining technique. It is the unsupervised classification of patterns into groups. Many algorithms for large data sets have been proposed in the literature using different techniques. However, conventional algorithms have some shortcomings such as slowness of the convergence, sensitive to initial value and preset classed in large scale data set etc. and they still require much investigation to improve performance and efficiency. Over the last decade, clustering with ant-based and swarm-based algorithms are emerging as an alternative to more traditional clustering techniques. Many complex optimization problems still exist, and it is often very difficult to obtain the desired result with one of these algorithms alone. Thus, robust and flexible techniques of optimization are needed to generate good results for clustering data. Some algorithms that imitate certain natural principles, known as evolutionary algorithms have been used in a wide variety of real-world applications. Recently, much research has been proposed using hybrid evolutionary algorithms to solve the clustering problem. This paper provides a survey of hybrid evolutionary algorithms for cluster analysis.

A new Hierarchical Pattern Recognition method using Mirroring Neural Networks

this paper, we develop a hierarchical classifier (an inverted tree-like structure) consisting of an organized set of each of which is actually a module that performs a feature extraction and an associated classification. We build each of such blocks by coupling a Mirroring Neural Network (MNN) with a clustering (algorithm) wherein the functions of the MNN are automatic data reduction and feature extraction which precedes an unsupervised classification. We then device an algorithm which we name as a Tandem Algorithm for the self-supervised learning of the MNN and an ensuing process of unsupervised pattern classification so that an ensemble of samples presented to the hierarchical classifier is classified and then sub-classified automatically. This tandem process is a two step process (feature extraction/data reduction and classification), implemented at each block (module) and can be extended level by level in the hierarchical architecture. The proposed procedure is practically demonstrated using 2 example cases where in a collage of images consisting of faces, flowers and furniture are classified and sub classified automatically. We expect that this kind of architecture will be very useful in the development of efficient and powerful self-learning machines in the future.