Unsupervised Mode Research Articles

Why is this an anomaly? Explaining anomalies using sequential explanations

In most applications, anomaly detection operates in an unsupervised mode by looking for outliers hoping that they are anomalies. Unfortunately, most anomaly detectors do not come with explanations about which features make a detected outlier point anomalous. Therefore, it requires human analysts to manually browse through each detected outlier point’s feature space to obtain the subset of features that will help them determine whether they are genuinely anomalous or not. This paper introduces sequential explanation (SE) methods that sequentially explain to the analyst which features make the detected outlier anomalous. We present two methods for computing SEs called the outlier and sample-based SE that will work alongside any anomaly detector. The outlier-based SE methods use an anomaly detector’s outlier scoring measure guided by a search algorithm to compute the SEs. Meanwhile, the sample-based SE methods employ sampling to turn the problem into a classical feature selection problem. In our experiments, we compare the performances of the different outlier- and sample-based SEs. Our results show that both the outlier and sample-based methods compute SEs that perform well and outperform sequential feature explanations.

End-to-End Deep One-Class Learning for Anomaly Detection in UAV Video Stream.

In recent years, the use of drones for surveillance tasks has been on the rise worldwide. However, in the context of anomaly detection, only normal events are available for the learning process. Therefore, the implementation of a generative learning method in an unsupervised mode to solve this problem becomes fundamental. In this context, we propose a new end-to-end architecture capable of generating optical flow images from original UAV images and extracting compact spatio-temporal characteristics for anomaly detection purposes. It is designed with a custom loss function as a sum of three terms, the reconstruction loss (), the generation loss () and the compactness loss () to ensure an efficient classification of the “deep-one” class. In addition, we propose to minimize the effect of UAV motion in video processing by applying background subtraction on optical flow images. We tested our method on very complex datasets called the mini-drone video dataset, and obtained results surpassing existing techniques’ performances with an AUC of 85.3.

A Fuzzy Rule-Based GIS Framework to Partition an Urban System Based on Characteristics of Urban Greenery in Relation to the Urban Context

We present a Geographical Information System (GIS)-based framework implementing a Mamdani fuzzy rule-based system to partition in an unsupervised mode an urban system in urban green areas. The proposed framework is characterized by high usability and flexibility. The study area is partitioned into homogeneous regions regarding the characteristics of public green areas and relations with the residents and buildings. The urban system is initially partitioned into microzones, given the smallest areas in which a census of the urban system is taken in terms of resident population, type and number of buildings and properties, and industrial and service activities. During a pre-processing phase, the values of specific indicators defined by a domain expert, which characterize the type of urban green area and the relationship with the residents and buildings, are calculated for each microzone. Subsequently, the fuzzy rule-based system component is executed to classify each microzone based on the fuzzy rule set constructed by the domain expert. Spatially adjoining microzones belonging to the same class are dissolved to form homogeneous areas called urban green contexts. The membership degrees of the microzones to the fuzzy set of their class are used to evaluate the reliability of the classification of the urban green context. We test our framework on the municipality of Pozzuoli, Italy, comparing the results with the ones obtained in a supervised manner by the expert appropriately partitioning and classifying the urban study area based on his knowledge of it.

High-Order Triplet CRF-PCANet for Unsupervised Segmentation of Nonstationary SAR Image

Conditional random fields (CRFs) model is suitable for image segmentation because it can capture the dependencies of observed data and incorporate the spatial correlations into the segmentation process. In this article, to deal with the segmentation of nonstationary synthetic aperture radar (SAR) image, we combine the modeling power of the CRF model with the representation-learning ability of principal component analysis network (PCANet), and thus propose a high-order triplet CRF model based on PCANet (HOTCRF-PCANet). HOTCRF-PCANet introduces an auxiliary field to explicitly regulate nonstationary label structure patterns. Under the guidance of this auxiliary field, HOTCRF-PCANet defines a discrete quadrilateral nonstationary Markov fields model, and thus considers both the nonstationary property of image and high-order label interactions. In addition, guided by the auxiliary field, HOTCRF-PCANet proposes to use a product-of-expert (POE) potential to enforce the regions’ labeling consistency for pixels within the weak-structured region. To automatically learn rich feature representations, HOTCRF-PCANet modifies PCANet into an unsupervised mode, i.e., unsupervised PCANet (UPCANet), and constructs an UPCANet-based unary potential to effectively predict the local class probability. The effectiveness of HOTCRF-PCANet is demonstrated by the application to the unsupervised segmentation of the simulated images and real SAR images.

A combined multivariate approach analyzing geochemical data for knowledge discovery: The Vazante – Paracatu Zinc District, Minas Gerais, Brazil

The Vazante Group comprises a Proterozoic sequence of carbonate and siliciclastic rocks located in Minas Gerais, Brazil. It is the host of the Vazante-Paracatu Zinc District, including world-class hypogene zinc silicate deposits, in the southern part, and several Pb – Zn sulfide deposits in the northern part, all hosted in dolomitic rocks. A recent study revealed the occurrence of base metal sulfide mineralization that formed prior to the Brasiliano orogenic event in the siliciclastic rocks (Serra do Garrote Formation) that underlie these dolomite-hosted silicates and sulfide deposits. These siliciclastic rocks were considered as potential sources of elements for the hydrothermal fluids that formed the dolomite-hosted deposits, however there was little evidence of depletion of the source rocks during the orogenic event. In this paper, Random Forest is used in unsupervised mode along with t-distributed Stochastic Neighbor Embedding (t-SNE) and principal component analysis (PCA) over a lithogeochemical dataset of samples of the Serra do Garrote siliciclastic rocks collected throughout the basin to provide insights about processes related to the mineralizing system at the Vazante-Paracatu District.Multivariate analysis reveals that the Serra do Garrote Formation geochemical signature typical of pre-orogenic mineralization. This is characterized by PC1+ with association of Zn, Cd, Cu, Hg, In, V ± Sb, Se, Mo, Re, which corresponds to nearly 60% of the variance in the data. This suggest that these elements are related to a widespread and/or, long-living hydrothermal activity without an efficient mechanism to focus the metal-bearing fluids, causing basin-scale sub-economical Zn and related metal enrichment. Furthermore, PC2 and PC6 distinguish the zones with signatures interpreted to be related to depletion. PC2+ separates Cd-poor sphalerite, which is typical of the generation of sphalerite that occur in zones with textural evidence of remobilization. PC6+ with association of As, organic carbon and S, identified the zones where sphalerite was leached from the pyrite-rich layers. Areas with multivariate signatures of both pre-orogenic enrichment and syn-orogenic depletion in the source siliciclastic rocks include the Vazante-North Extension and Varginha zinc silicate deposits/occurrence, Ambrósia silicate zinc deposit and the Engenho Velho prospect. Preliminary exploration in this prospect revealed hydrothermal alteration typical of the zinc silicate deposits. This study is pioneering in applying these numerical models in possible source rocks to assist in identifying targets for exploration in basins.

Commemorative insights: the best of life, in death

ABSTRACT This study explores visitor experience at the National Anzac Centre in Western Australia using multiple qualitative methods. Initially, Nethnography is used to assemble a blend of lived experience and online non-dialogical commentary. Nethnography (an alternative to Netnography) is used here as a mechanism for data grooming. Three data sets inform this study: 500 Trip Advisor comments, 500 Visitors’ Book comments and four days of participant observation. The data are then analysed using Reflexive Thematic Analysis and Leximancer in the unsupervised mode. This methodological collage is designed to improve the veracity of interpretation through both lived experience and triangulation across data sources. Findings suggest a significant visitor-thirst for the positive aspects of commemoration. By the same token no respondent reported being motivated by schadenfreude, mortality salience or death. If a certain fascination with, and commodification of death defines popular dark tourism then commemorative tourism’s relegation of death indicates exception. It would seem commemorists relegate death and darkness to mere context, while gravitas, ritual and cultural validation transcend the superficial and the kitsch. Meanwhile, visitors to the National Anzac Centre concentrate on more endearing traits including sacrifice, love, loss and the nobility of caritas.

Web-based and mixed-mode cognitive large-scale assessments in higher education: An evaluation of selection bias, measurement bias, and prediction bias

Educational large-scale studies typically adopt highly standardized settings to collect cognitive data on large samples of respondents. Increasing costs alongside dwindling response rates in these studies necessitate exploring alternative assessment strategies such as unsupervised web-based testing. Before respective assessment modes can be implemented on a broad scale, their impact on cognitive measurements needs to be quantified. Therefore, an experimental study on N = 17,473 university students from the German National Educational Panel Study has been conducted. Respondents were randomly assigned to a supervised paper-based, a supervised computerized, and an unsupervised web-based mode to work on a test of scientific literacy. Mode-specific effects on selection bias, measurement bias, and predictive bias were examined. The results showed a higher response rate in web-based testing as compared to the supervised modes, without introducing a pronounced mode-specific selection bias. Analyses of differential test functioning showed systematically larger test scores in paper-based testing, particularly among low to medium ability respondents. Prediction bias for web-based testing was observed for one out of four criteria on study-related success factors. Overall, the results indicate that unsupervised web-based testing is not strictly equivalent to other assessment modes. However, the respective bias introduced by web-based testing was generally small. Thus, unsupervised web-based assessments seem to be a feasible option in cognitive large-scale studies in higher education.

Optimal probabilistic catalogue matching for radio sources

ABSTRACT Cross-matching catalogues from radio surveys to catalogues of sources at other wavelengths is extremely hard, because radio sources are often extended, often consist of several spatially separated components, and often no radio component is coincident with the optical/infrared host galaxy. Traditionally, the cross-matching is done by eye, but this does not scale to the millions of radio sources expected from the next generation of radio surveys. We present an innovative automated procedure, using Bayesian hypothesis testing, that models trial radio-source morphologies with putative positions of the host galaxy. This new algorithm differs from an earlier version by allowing more complex radio-source morphologies, and performing a simultaneous fit over a large field. We show that this technique performs well in an unsupervised mode.

Transfer learning aid the prediction of sintering densification

In powder metallurgy engineering, the master sintering curve (MSC) is crucial for estimating the mechanical properties of sintered products and optimizing sintering process parameters. A rapid evaluation method, that is, a domain-adversarial neural network, is established in the field of sintering to transfer learning from Al2O3 to SiC, concerning the effect of the heating rate on the sintering densification, in which one material (e.g. SiC) lacks an MSC and the material (e.g. Al2O3) has an overabundance of MSCs. In the unsupervised mode, we can roughly predict the densification evolution of SiC based on the MSC data of Al2O3. In the semi-supervised mode, prediction accuracy gradually increases with the increase of Al2O3 data and approaches a certain upper limit. Compared with the traditional sintering density prediction models, the proposed approach can provide an effective and rapid solution to the problem of data scarcity in the sintering field.

GeneMark-EP+: eukaryotic gene prediction with self-training in the space of genes and proteins.

We have made several steps toward creating a fast and accurate algorithm for gene prediction in eukaryotic genomes. First, we introduced an automated method for efficient ab initio gene finding, GeneMark-ES, with parameters trained in iterative unsupervised mode. Next, in GeneMark-ET we proposed a method of integration of unsupervised training with information on intron positions revealed by mapping short RNA reads. Now we describe GeneMark-EP, a tool that utilizes another source of external information, a protein database, readily available prior to the start of a sequencing project. A new specialized pipeline, ProtHint, initiates massive protein mapping to genome and extracts hints to splice sites and translation start and stop sites of potential genes. GeneMark-EP uses the hints to improve estimation of model parameters as well as to adjust coordinates of predicted genes if they disagree with the most reliable hints (the -EP+ mode). Tests of GeneMark-EP and -EP+ demonstrated improvements in gene prediction accuracy in comparison with GeneMark-ES, while the GeneMark-EP+ showed higher accuracy than GeneMark-ET. We have observed that the most pronounced improvements in gene prediction accuracy happened in large eukaryotic genomes.

Anomaly detection of defects on concrete structures with the convolutional autoencoder

This paper reports the application of deep learning for implementing the anomaly detection of defects on concrete structures, so as to facilitate the visual inspection of civil infrastructure. A convolutional autoencoder was trained as a reconstruction-based model, with the defect-free images, to rapidly and reliably detect defects from the large volume of image datasets. This training process was in the unsupervised mode, with no label needed, thereby requiring no prior knowledge and saving an enormous amount of time for label preparation. The built anomaly detector favors minimizing the reconstruction errors of defect-free images, which renders high reconstruction errors of defects, in turn, detecting the location of defects. The assessment shows that the proposed anomaly detection technique is robust and adaptable to defects on wide ranges of scales. Comparison was also made with the segmentation results produced by other automatic classical methods, revealing that the results made by the anomaly map outperform other segmentation methods, in terms of precision, recall, F1 measure and F2 measure, without severe under- and over-segmentation. Further, instead of merely being a binary map, each pixel of the anomaly map is represented by the anomaly score, which acts as a risk indicator for alerting inspectors, wherever defects on concrete structures are detected.

A new unsupervised method for boundary perception and word-like segmentation of sequence

In cognitive science research on natural language processing, motor learning and visual perception, perceiving boundary points and segmenting a continuous string or sequence is one of the fundamental problems. Boundary perception can also be viewed as a machine learning problem; supervised or unsupervised learning. In supervised learning approach for determining boundary points for segmentation of a sequence, it is necessary to have some pre-segmented training examples. In unsupervised mode, the learning is accomplished without any training data hence, the frequency of occurence of symbols within the sequence is normally used as the cue. Most of earlier algorithms use this cue while scanning the sequence in forward direction. In this paper we propose a novel approach of extracting the possible boundary points by using bi-directional scanning of the sequence. We show here that such an extension from unidirectional to bi-directional is not trivial and requires judicious consideration of datastructure and algorithm. We here propose a new algorithm which traverses the sequence unidirectionally but extracts the information bi-directionally. Our method yields better segmentation which is demonstrated by rigorous experimentation on several datasets.

An Unsupervised Embedding Learning Feature Representation Scheme for Network Big Data Analysis

With the arrival of the big data era, data come frequently with increasing volume and high dimensionality, which imposes a considerable challenge on data compression in network representation and analysis. How to learn an effective low-dimensional representation has a dramatic influence on performance of specific network learning tasks. In this paper, we propose an unsupervised embedding learning feature representation scheme by deep Siamese neural networks, aiming to learn an efficient low-dimensional feature subspace. Unsupervised embedding learning is one tough but interesting task in which its searching strategy is performed without the guidance of class label information. Siamese network is a neural network that can learn an efficient feature subspace in a supervised mode. It trains two networks with shared weights simultaneously, and feeds them with random sampling from the same dataset. As a result, the feature space is projected onto a low-dimensional subspace such that the similar samples are with small values close to zero, whereas the dissimilar ones come with big values greater than a predefined margin. We further discuss deep Siamese neural network in an unsupervised mode and its applications to embedding learning. The proposed method can be also used to address semi-supervised feature representation problems. Finally, the learned unsupervised embedding is validated on eight publicly available databases including images, voices, and text documents. Extensive experiments demonstrate the superiority of the proposed method against the compared existing state-of-the-art embedding approaches.

A new unsupervised method for boundary perception and word-like segmentation of sequence

In cognitive science research on natural language processing, motor learning and visual perception, perceiving boundary points and segmenting a continuous string or sequence is one of the fundamental problems. Boundary perception can also be viewed as a machine learning problem; supervised or unsupervised learning. In supervised learning approach for determining boundary points for segmentation of a sequence, it is necessary to have some pre-segmented training examples. In unsupervised mode, the learning is accomplished without any training data hence, the frequency of occurence of symbols within the sequence is normally used as the cue. Most of earlier algorithms use this cue while scanning the sequence in forward direction. In this paper we propose a novel approach of extracting the possible boundary points by using bi-directional scanning of the sequence. We show here that such an extension from unidirectional to bi-directional is not trivial and requires judicious consideration of datastructure and algorithm. We here propose a new algorithm which traverses the sequence unidirectionally but extracts the information bi-directionally. Our method yields better segmentation which is demonstrated by rigorous experimentation on several datasets.

Spike time dependent plasticity (STDP) enabled learning in spiking neural networks using domain wall based synapses and neurons

We have implemented a Spiking Neural Network (SNN) architecture using a combination of spin orbit torque driven domain wall devices and transistor based peripheral circuits as both synapses and neurons. Learning in the SNN hardware is achieved both under completely unsupervised mode and partially supervised mode through mechanisms, incorporated in our spintronic synapses and neurons, that have biological plausibility, e.g., Spike Time Dependent Plasticity (STDP) and homoeostasis. High classification accuracy is obtained on the popular Iris dataset for both modes of learning.

In quest of effect directed analysis in the smart laboratory: Automated system for flow-through evaluation of membranotropic effects of emerging contaminants

The rate-determining step of the human exposome workflow is the acquisition of physiologically relevant data (e.g., effect directed analysis), which can be performed retrospectively or with ad hoc experiments. In this contribution, an automated system is proposed for evaluating potential interaction mechanisms of xenobiotics across cell membranes, the so-called membranotropic effects, using liposomes as a mimicry of biological membranes, and fluorescent membrane probes. The smart fluidic method features real-time acquisition of fluorescence readouts, data processing and feedback in a fully unsupervised mode. As a proof of concept applicability, the behavior of newly synthesized cholesterol-laden biomimetic liposomes, and the in-vitro potential toxicant action of bisphenol A and diclofenac as model of emerging contaminants on cell membrane surrogates were investigated in a flow-through format. Unattended operation resulted in excellent intermediate precision (<1.5%) and unveiled that diclofenac affected the liposomal bilayer order very slightly, regardless of the cholesterol concentration, because it accumulates at a superficial level, while the membranotropic effect of bisphenol A was more pronounced at low concentration levels of cholesterol because at increased levels, the membrane reduces its permeability.

Unsupervised Spatial–Spectral Feature Learning by 3D Convolutional Autoencoder for Hyperspectral Classification

Feature learning technologies using convolutional neural networks (CNNs) have shown superior performance over traditional hand-crafted feature extraction algorithms. However, a large number of labeled samples are generally required for CNN to learn effective features under classification task, which are hard to be obtained for hyperspectral remote sensing images. Therefore, in this paper, an unsupervised spatial–spectral feature learning strategy is proposed for hyperspectral images using 3-Dimensional (3D) convolutional autoencoder (3D-CAE). The proposed 3D-CAE consists of 3D or elementwise operations only, such as 3D convolution, 3D pooling, and 3D batch normalization, to maximally explore spatial–spectral structure information for feature extraction. A companion 3D convolutional decoder network is also designed to reconstruct the input patterns to the proposed 3D-CAE, by which all the parameters involved in the network can be trained without labeled training samples. As a result, effective features are learned in an unsupervised mode that label information of pixels is not required. Experimental results on several benchmark hyperspectral data sets have demonstrated that our proposed 3D-CAE is very effective in extracting spatial–spectral features and outperforms not only traditional unsupervised feature extraction algorithms but also many supervised feature extraction algorithms in classification application.

Automatic Mesofluidic System Combining Dynamic Gastrointestinal Bioaccessibility with Lab-on-Valve-Based Sorptive Microextraction for Risk Exposure of Organic Emerging Contaminants in Filter-Feeding Organisms.

An automatic mesofluidic system combining dynamic oral bioaccessibility with lab-on-valve (LOV)-based sorptive microextraction is herein proposed for the first time for exploring the kinetics of leaching of incurred rather than spiked organic emerging contaminants (viz., methyl paraben, butyl paraben, diclofenac, and triclosan) from exposed mussels on the basis of the Versantvoort's fed-state physiological extraction test. Our method capitalizes on programmable flow analysis, in which gastrointestinal extracts are obtained online by pumping a simulated biorelevant gastrointestinal fluid across a large-bore column (maintained at 37.0 ± 2.0 °C) loaded with 250 mg of freeze-dried and powdered mussel onto a polyvinyldiene difluoride filter membrane. The physiologically relevant extracts are then cleaned up, and the analytes are preconcentrated onto a dedicated reversed-phase solid-phase extraction (Oasis PRIME-HLB) microcolumn that is captured into the channels of an LOV mesofluidic platform. The aim behind this is to obtain analyte-laden eluates with ACN/MeOH (90:10, v/v) in unsupervised mode for direct injection into LC-MS. The LOV minicolumn (≤25 mg) is automatically disposed of and renewed for every individual fraction on account of the strong retention of (phospho)lipids by the copolymeric sorbent. The proposed dynamic bioaccessibility test features a significant shortening of the extraction time against the batch method (28 vs 240 min) while avoiding overestimation of potentially bioavailable fractions. The trueness of the online gastrointestinal extraction method was confirmed using mass-balance validation following ultrasonic-assisted solid-liquid extraction of the original mussel sample and the residual (nonbioaccessible) fractionof emerging contaminants.

Improving fuzzy c-means clustering via quantum-enhanced weighted superposition attraction algorithm

Fuzzy clustering has become an important research field in pattern recognition and data analysis. As supporting unsupervised mode of learning, fuzzy clustering brings about unique opportunities to reveal structural relationships in data. Fuzzy c-means clustering is one of the widely preferred clustering algorithms in the literature. However, fuzzy c-means clustering algorithm has a major drawback that it can get trapped at some local optima. In order to overcome this shortcoming, this study employs a new generation metaheuristic algorithm. Weighted Superposition Attraction Algorithm (WSA) is a novel swarm intelligence-based method that draws inspiration from the superposition principle of physics in combination with the attracted movement of agents. Due to its high converging capability and practicality, WSA algorithm has been employed in order to enhance performance of fuzzy-c means clustering. Comprehensive experimental study has been conducted on publicly available datasets obtained from UCI machine learning repository. The results point out significant improvements over the traditional fuzzy c-means algorithm.

Granular Mining and Big Data Analytics: Rough Models and Challenges

Data analytics in granular computing framework is considered for several mining applications, such as in video analysis, bioinformatics and online social networks which have all the characteristics of Big data. The role of granulation, lower approximation and r–f information measure is exhibited. While the lower approximation over a video sequence signifies the object model for unsupervised tracking, it characterizes the probability (relative frequency) of definite regions in ranking miRNAs for normal and cancer classification. For neural learning, the information on definite region is used as the initial knowledge for encoding while generating the networks through evolution. Granules considered are of different sizes and dimensions with fuzzy and crisp boundaries. The tracking method is effective in handling different ambiguous situations, e.g., overlapping objects, newly appeared object(s), multiple objects in different directions and speeds, in unsupervised mode. The ranking algorithm could find only 1% miRNAs to result in significantly higher F-score than the entire set. Fuzzy–rough communities detected over the granular model of social networks are suitable in dealing with overlapping virtual communities in Big data. The knowledge encoding based on fuzzy–rough set provides superior performance than that of rough set. Future directions of research and challenges including the significance of z-numbers in precisiation of granules are stated. The article includes some of the results published elsewhere.