Supervised Clustering Approach Research Articles

Supervised Clustering of Persian Handwritten Images Using Regularization and Dimension Reduction Methods

Clustering, as a fundamental exploratory data technique, not only is used to discover patterns and structures in complex datasets but also is utilized to group variables in high-dimensional data analysis. Dimension reduction through clustering helps identify important variables and reduce data dimensions without losing significant information. High-dimensional image datasets, such as Persian handwritten images, have numerous pixels, making statistical inference difficult. Such high-dimensionality property pose challenges for analysis and processing, requiring specialized techniques like clustering to extract information. Incorporating response variable information enhances clustering analysis, transforming it into a supervised method. This article evaluates a supervised clustering approach using Ridge and Lasso penalties, comparing them in analyzing a real dataset while identifying important variables. We demonstrate that despite choosing a small number of variables as important variables, Lasso penalty performs relatively well in predicting the labels of new observations for this multi-class dataset.

Derivation of Maternal Dietary Patterns Accounting for Regional Heterogeneity

Abstract Latent class models are often used to characterise dietary patterns. Yet, when subtle variations exist across different sub-populations, overall population patterns can be masked and affect statistical inference on health outcomes. We address this concern with a flexible supervised clustering approach, introduced as Supervised Robust Profile Clustering, that identifies outcome-dependent population-based patterns, while partitioning out subpopulation pattern differences. Using dietary data from the 1997–2011 National Birth Defects Prevention Study, we determine how maternal dietary profiles associate with orofacial clefts among offspring. Results indicate mothers who consume a higher proportion of fruits and vegetables compared to land meats lower the proportion of progeny with orofacial cleft defect.

Accurate Molecular-Orbital-Based Machine Learning Energies via Unsupervised Clustering of Chemical Space.

We introduce an unsupervised clustering algorithm to improve training efficiency and accuracy in predicting energies using molecular-orbital-based machine learning (MOB-ML). This work determines clusters via the Gaussian mixture model (GMM) in an entirely automatic manner and simplifies an earlier supervised clustering approach [ J. Chem. Theory Comput. 2019, 15, 6668] by eliminating both the necessity for user-specified parameters and the training of an additional classifier. Unsupervised clustering results from GMM have the advantages of accurately reproducing chemically intuitive groupings of frontier molecular orbitals and exhibiting improved performance with an increasing number of training examples. The resulting clusters from supervised or unsupervised clustering are further combined with scalable Gaussian process regression (GPR) or linear regression (LR) to learn molecular energies accurately by generating a local regression model in each cluster. Among all four combinations of regressors and clustering methods, GMM combined with scalable exact GPR (GMM/GPR) is the most efficient training protocol for MOB-ML. The numerical tests of molecular energy learning on thermalized data sets of drug-like molecules demonstrate the improved accuracy, transferability, and learning efficiency of GMM/GPR over other training protocols for MOB-ML, i.e., supervised regression clustering combined with GPR (RC/GPR) and GPR without clustering. GMM/GPR also provides the best molecular energy predictions compared with ones from the literature on the same benchmark data sets. With a lower scaling, GMM/GPR has a 10.4-fold speedup in wall-clock training time compared with scalable exact GPR with a training size of 6500 QM7b-T molecules.

A Novel Autochthonous Mouse Model Serves As a Preclinical Evaluation Platform and Explores Dual BTK and BCL2 Inhibition for Activated B Cell-like Diffuse Large B Cell Lymphoma

A Novel Autochthonous Mouse Model Serves As a Preclinical Evaluation Platform and Explores Dual BTK and BCL2 Inhibition for Activated B Cell-like Diffuse Large B Cell Lymphoma

Phenotypic profile clustering pragmatically identifies diagnostically and mechanistically informative subgroups of chronic pain patients.

Traditional classification and prognostic approaches for chronic pain conditions focus primarily on anatomically based clinical characteristics not based on underlying biopsychosocial factors contributing to perception of clinical pain and future pain trajectories. Using a supervised clustering approach in a cohort of temporomandibular disorder cases and controls from the Orofacial Pain: Prospective Evaluation and Risk Assessment study, we recently developed and validated a rapid algorithm (ROPA) to pragmatically classify chronic pain patients into 3 groups that differed in clinical pain report, biopsychosocial profiles, functional limitations, and comorbid conditions. The present aim was to examine the generalizability of this clustering procedure in 2 additional cohorts: a cohort of patients with chronic overlapping pain conditions (Complex Persistent Pain Conditions study) and a real-world clinical population of patients seeking treatment at duke innovative pain therapies. In each cohort, we applied a ROPA for cluster prediction, which requires only 4 input variables: pressure pain threshold and anxiety, depression, and somatization scales. In both complex persistent pain condition and duke innovative pain therapies, we distinguished 3 clusters, including one with more severe clinical characteristics and psychological distress. We observed strong concordance with observed cluster solutions, indicating the ROPA method allows for reliable subtyping of clinical populations with minimal patient burden. The ROPA clustering algorithm represents a rapid and valid stratification tool independent of anatomic diagnosis. ROPA holds promise in classifying patients based on pathophysiological mechanisms rather than structural or anatomical diagnoses. As such, this method of classifying patients will facilitate personalized pain medicine for patients with chronic pain.

Deep prediction of investor interest: A supervised clustering approach

We propose a novel deep learning architecture suitable for the prediction of investor interest for a given asset in a given time frame. This architecture performs both investor clustering and modelling at the same time. We first verify its superior performance on a synthetic scenario inspired by real data and then apply it to two real-world databases, a publicly available dataset about the position of investors in Spanish stock market and proprietary data from BNP Paribas Corporate and Institutional Banking.1,2

Analysis of hyperspectral images for detection of drought stress and recovery in maize plants in a high-throughput phenotyping platform

The study of physiological processes resulting from water-limited conditions in crops is essential for the selection of drought-tolerant genotypes and the functional analysis of related genes. A promising, non-invasive technique for plant trait analysis is close-range hyperspectral imaging (HSI), which has great potential for the early detection of plant responses to water deficit stress. In this work, a data analysis method is described that, unlike vegetation indices, the present method applies spectral similarity on selected bands with high discriminative information, while requiring a careful treatment of uninformative illumination effects. The latter issue is solved by a standard normal variate (SNV) normalization that removes linear effects and a supervised clustering approach to remove pixels that exhibit nonlinear multiple scattering effects. On the remaining pixels, the stress-related dynamics is quantified by a spectral analysis procedure that involves a supervised band selection procedure and a spectral similarity measure against well-watered control plants. The proposed method was validated by a large-scale study of water-stress and recovery of maize plants in a high-throughput plant phenotyping platform. The results showed that the analysis method allows for an early detection of drought stress responses and of recovery effects shortly after re-watering.

Microstructure Cluster Analysis with Transfer Learning and Unsupervised Learning

We apply computer vision and machine learning methods to analyze two datasets of microstructural images. A transfer learning pipeline utilizes the fully connected layer of a pre-trained convolutional neural network as the image representation. An unsupervised learning method uses the image representations to discover visually distinct clusters of images within two datasets. A minimally supervised clustering approach classifies micrographs into visually similar groups. This approach successfully classifies images both in a dataset of surface defects in steel, where the image classes are visually distinct and in a dataset of fracture surfaces that humans have difficulty classifying. We find that the unsupervised, transfer learning method gives results comparable to fully supervised, custom-built approaches.

Taxonomy matching between asteroids and meteorites: supervised clustering approach

We propose a data-driven method for investigating taxonomy relations between asteroids and meteorite. The method clusters data of reflectance spectra of asteroids with the guidance of relatively reliable taxonomy of meteorites. By comparing the clustering accuracy of asteroid between with and without guidance, we observe that the guidance of meteorites taxonomy improve the accuracy. This serves as an evidence that there is a common taxonomical structure between meteorites and asteroids, confirming a long-standing prediction.

Learning Facial Action Units from Web Images with Scalable Weakly Supervised Clustering.

We present a scalable weakly supervised clustering approach to learn facial action units (AUs) from large, freely available web images. Unlike most existing methods (e.g., CNNs) that rely on fully annotated data, our method exploits web images with inaccurate annotations. Specifically, we derive a weakly-supervised spectral algorithm that learns an embedding space to couple image appearance and semantics. The algorithm has efficient gradient update, and scales up to large quantities of images with a stochastic extension. With the learned embedding space, we adopt rank-order clustering to identify groups of visually and semantically similar images, and re-annotate these groups for training AU classifiers. Evaluation on the 1 millon EmotioNet dataset demonstrates the effectiveness of our approach: (1) our learned annotations reach on average 91.3% agreement with human annotations on 7 common AUs, (2) classifiers trained with re-annotated images perform comparably to, sometimes even better than, its supervised CNN-based counterpart, and (3) our method offers intuitive outlier/noise pruning instead of forcing one annotation to every image. Code is available.

3D modeling of reservoir electrofacies using integration clustering and geostatistic method in central field of Persian Gulf

Description and distribution of reservoir rock characterization in 3D space are challenging issues in study of heterogeneous reservoirs. In order to obviate these complications, a static model based on reservoir and geology characterization is created. Well log data has been using to extract reservoir properties. Analyzing a large volume of well log data in order to extract reservoir properties by manual approaches is difficult and time consuming. Therefore, reservoir properties can be determined by simultaneous process of several logs. The simultaneous process to classify well log data based on similar response is called electrofacies (EF). Electrofacies analysis is used to determine reservoir properties based upon lithology, porosity and permeability. This study consists of three steps. Initially, based on similarity in well log response and geological characteristics, electrofacies were created by unsupervised clustering approach in one well. These approaches include hierarchical cluster analysis (HCA), multi resolution graph based clustering (MRGC) and Self-Organizing Maps (SOM). Afterward, the clusters were generalized to all wells by supervised clustering approach. Finally, geostatistical simulation was applied to generate a 3D spatial model of reservoir electrofacies. According to the results, the hierarchical clustering has performed as a robust and more effective approach in clustering of data based on silhouette validity tests and geological information. Application of sequential indicator simulation (SIS) algorithms to estimate EF 3D model was successfully tested in the study area.

Prognostic relevance of gene signatures in high-grade serous ovarian carcinoma.

5510 Background: Transcriptional profiling of ovarian cancers has proven to be complex. As such it has been difficult to validate existing signatures across studies. Cancer Genome Atlas (TCGA) researchers have identified four molecular subtypes of high-grade serous ovarian cancer (HGSOC). However, survival duration did not differ significantly for the TCGA subtypes. Potential limitations of the TCGA data include short clinical follow-up (45% were alive at the time of last follow-up) and the need to unify gene expression measures from multiple platforms. Methods: Clinically annotated stage-II–IV HGSOC samples (n = 175) with >70% tumor cell content were profiled using the Agilent Whole Human Genome 4 x 44K chip. To identify subtypes non-negative matrix factorization (NMF) of mRNA expression was performed using ~2000 genes with the highest variability across patients. In parallel differentially expressed genes were identified using the Rosetta Similarity Search Tool as well as analysis of variance based on genes known to be involved in epithelial to mesenchymal transition. Results: Median follow-up time was 35 months (range, 0–202 months, 12% were alive at the time of last follow-up). NMF clustering confirmed four HGSOC subtypes (immunoreactive, differentiated, proliferative, and mesenchymal) on the basis of gene content in the clusters. Pathway signatures with therapeutic potential were identified for individual or multiple subtypes. Survival differed significantly between the four molecular subgroups in univariate (Hazard Ratio [HR] 2.4, 95% CI 1.5-4.1, p = 0.007) and multivariate (HR 2.3, 95% CI 1.3-4.0, p = 0.003) analyses when accounting for age, stage, grade, and postoperative residual tumor. Using the supervised clustering approach two distinct molecular subtypes of HGSOC based on epithelial and mesenchymal gene expression signatures were identified with significantly different survival outcomes. Conclusions: Here we independently validate and expand upon the molecular TCGA classification of HGSOC. The potential of these two prognostic classifiers may lie in their ability to recognize categories of patients that are more likely to respond to particular therapies.

A supervised clustering approach for fMRI-based inference of brain states

We propose a method that combines signals from many brain regions observed in functional Magnetic Resonance Imaging (fMRI) to predict the subject's behavior during a scanning session. Such predictions suffer from the huge number of brain regions sampled on the voxel grid of standard fMRI data sets: the curse of dimensionality. Dimensionality reduction is thus needed, but it is often performed using a univariate feature selection procedure, that handles neither the spatial structure of the images, nor the multivariate nature of the signal. By introducing a hierarchical clustering of the brain volume that incorporates connectivity constraints, we reduce the span of the possible spatial configurations to a single tree of nested regions tailored to the signal. We then prune the tree in a supervised setting, hence the name supervised clustering, in order to extract a parcellation (division of the volume) such that parcel-based signal averages best predict the target information. Dimensionality reduction is thus achieved by feature agglomeration, and the constructed features now provide a multi-scale representation of the signal. Comparisons with reference methods on both simulated and real data show that our approach yields higher prediction accuracy than standard voxel-based approaches. Moreover, the method infers an explicit weighting of the regions involved in the regression or classification task.

Use of linguistic petrographical descriptions to characterise core porosity: contrasting approaches

There are many classification problems in petroleum reservoir characterisation, an example being the recognition of lithofacies from well log data. Data classification is not an easy task when the data are not of numerical origin. This paper compares three approaches to classify porosity into groups (very poor, poor, fair, good) using petrographical characteristics described in linguistic terms. The three techniques used are an expert system approach, a supervised clustering approach, and a neural network approach. From the results applied to a core data set in Australia, we found that the techniques performed best in decreasing order of their requirement for significant user effort, for a low degree of benefit achieved thereby.