Sparse Classification Research Articles

Decoding memory features from hippocampal spiking activities using sparse classification models.

To understand how memory information is encoded in the hippocampus, we build classification models to decode memory features from hippocampal CA3 and CA1 spatio-temporal patterns of spikes recorded from epilepsy patients performing a memory-dependent delayed match-to-sample task. The classification model consists of a set of B-spline basis functions for extracting memory features from the spike patterns, and a sparse logistic regression classifier for generating binary categorical output of memory features. Results show that classification models can extract significant amount of memory information with respects to types of memory tasks and categories of sample images used in the task, despite the high level of variability in prediction accuracy due to the small sample size. These results support the hypothesis that memories are encoded in the hippocampal activities and have important implication to the development of hippocampal memory prostheses.

Conducting sparse feature selection on arbitrarily long phrases in text corpora with a focus on interpretability

We propose a general framework for topic‐specific summarization of large text corpora, and illustrate how it can be used for analysis in two quite different contexts: an Occupational Safety and Health Administration (OSHA) database of fatality and catastrophe reports (to facilitate surveillance for patterns in circumstances leading to injury or death), and legal decisions on workers' compensation claims (to explore relevant case law). Our summarization framework, built on sparse classification methods, is a compromise between simple word frequency‐based methods currently in wide use, and more heavyweight, model‐intensive methods such as latent Dirichlet allocation (LDA). For a particular topic of interest (e.g., mental health disability, or carbon monoxide exposure), we regress a labeling of documents onto the high‐dimensional counts of all the other words and phrases in the documents. The resulting small set of phrases found as predictive are then harvested as the summary. Using a branch‐and‐bound approach, this method can incorporate phrases of arbitrary length, which allows for potentially rich summarization. We discuss how focus on the purpose of the summaries can inform choices of tuning parameters and model constraints. We evaluate this tool by comparing the computational time and summary statistics of the resulting word lists to three other methods in the literature. We also present a new R package, textreg. Overall, we argue that sparse methods have much to offer in text analysis and is a branch of research that should be considered further in this context. © 2016 Wiley Periodicals, Inc. Statistical Analysis and Data Mining: The ASA Data Science Journal, 2016

An efficient classification method based on principal component and sparse representation.

As an important application in optical imaging, palmprint recognition is interfered by many unfavorable factors. An effective fusion of blockwise bi-directional two-dimensional principal component analysis and grouping sparse classification is presented. The dimension reduction and normalizing are implemented by the blockwise bi-directional two-dimensional principal component analysis for palmprint images to extract feature matrixes, which are assembled into an overcomplete dictionary in sparse classification. A subspace orthogonal matching pursuit algorithm is designed to solve the grouping sparse representation. Finally, the classification result is gained by comparing the residual between testing and reconstructed images. Experiments are carried out on a palmprint database, and the results show that this method has better robustness against position and illumination changes of palmprint images, and can get higher rate of palmprint recognition.

A molecular classification of human mesenchymal stromal cells.

Mesenchymal stromal cells (MSC) are widely used for the study of mesenchymal tissue repair, and increasingly adopted for cell therapy, despite the lack of consensus on the identity of these cells. In part this is due to the lack of specificity of MSC markers. Distinguishing MSC from other stromal cells such as fibroblasts is particularly difficult using standard analysis of surface proteins, and there is an urgent need for improved classification approaches. Transcriptome profiling is commonly used to describe and compare different cell types; however, efforts to identify specific markers of rare cellular subsets may be confounded by the small sample sizes of most studies. Consequently, it is difficult to derive reproducible, and therefore useful markers. We addressed the question of MSC classification with a large integrative analysis of many public MSC datasets. We derived a sparse classifier (The Rohart MSC test) that accurately distinguished MSC from non-MSC samples with >97% accuracy on an internal training set of 635 samples from 41 studies derived on 10 different microarray platforms. The classifier was validated on an external test set of 1,291 samples from 65 studies derived on 15 different platforms, with >95% accuracy. The genes that contribute to the MSC classifier formed a protein-interaction network that included known MSC markers. Further evidence of the relevance of this new MSC panel came from the high number of Mendelian disorders associated with mutations in more than 65% of the network. These result in mesenchymal defects, particularly impacting on skeletal growth and function. The Rohart MSC test is a simple in silico test that accurately discriminates MSC from fibroblasts, other adult stem/progenitor cell types or differentiated stromal cells. It has been implemented in the www.stemformatics.org resource, to assist researchers wishing to benchmark their own MSC datasets or data from the public domain. The code is available from the CRAN repository and all data used to generate the MSC test is available to download via the Gene Expression Omnibus or the Stemformatics resource.

Action-Based Pedestrian Identification via Hierarchical Matching Pursuit and Order Preserving Sparse Coding

Pedestrian identification is a very important topic in the area of intelligent surveillance and public safety, where the near front face images of pedestrian can hardly be obtained due to high installation angle of camera, long-distance location and extreme light variations. This paper presents a new action-based pedestrian identification algorithm, which adopts hierarchical matching pursuit (HMP) to extract features and order preserving sparse coding (OPSC) to do classification. Two-layer HMP features are extracted from foreground frame image patches by sparse coding, max pooling and normalization, which preserve both local and global information. OPSC is taken as classifier to take full advantage of the spatial structure information, which is different from traditional temporal OPSC algorithm. The spatiotemporal order preserving sparse coding-based classification is also investigated. The effectiveness of the proposed method is verified on public data sets, and the experimental results show the superiority of our method.

Robust Fringe Projection Profilometry via Sparse Representation.

In this paper, a robust fringe projection profilometry (FPP) algorithm using the sparse dictionary learning and sparse coding techniques is proposed. When reconstructing the 3D model of objects, traditional FPP systems often fail to perform if the captured fringe images have a complex scene, such as having multiple and occluded objects. It introduces great difficulty to the phase unwrapping process of an FPP system that can result in serious distortion in the final reconstructed 3D model. For the proposed algorithm, it encodes the period order information, which is essential to phase unwrapping, into some texture patterns and embeds them to the projected fringe patterns. When the encoded fringe image is captured, a modified morphological component analysis and a sparse classification procedure are performed to decode and identify the embedded period order information. It is then used to assist the phase unwrapping process to deal with the different artifacts in the fringe images. Experimental results show that the proposed algorithm can significantly improve the robustness of an FPP system. It performs equally well no matter the fringe images have a simple or complex scene, or are affected due to the ambient lighting of the working environment.

An online sparse estimation-based classification approach for real-time monitoring in advanced manufacturing processes from heterogeneous sensor data

ABSTRACTThe objective of this work is to realize real-time monitoring of process conditions in advanced manufacturing using multiple heterogeneous sensor signals. To achieve this objective we propose an approach invoking the concept of sparse estimation called online sparse estimation-based classification (OSEC). The novelty of the OSEC approach is in representing data from sensor signals as an underdetermined linear system of equations and subsequently solving the underdetermined linear system using a newly developed greedy Bayesian estimation method. We apply the OSEC approach to two advanced manufacturing scenarios, namely, a fused filament fabrication additive manufacturing process and an ultraprecision semiconductor chemical–mechanical planarization process. Using the proposed OSEC approach, process drifts are detected and classified with higher accuracy compared with popular machine learning techniques. Process drifts were detected and classified with a fidelity approaching 90% (F-score) using OSEC. In comparison, conventional signal analysis techniques—e.g., neural networks, support vector machines, quadratic discriminant analysis, naïve Bayes—were evaluated with F-score in the range of 40% to 70%.

Analysis of Allergenic Pollen by FTIR Microspectroscopy.

Fourier transform infrared (FTIR) spectroscopy is a powerful tool for the identification and characterization of pollen and spores. However, interpretation and multivariate analysis of infrared microscopy spectra of single pollen grains are hampered by Mie-type scattering. In this paper, we introduce a novel sampling setup for infrared microspectroscopy of pollens preventing strong Mie-type scattering. Pollen samples were embedded in a soft paraffin layer between two sheets of polyethylene foils without any further sample pretreatment. Single-grain infrared spectra of 13 different pollen samples, belonging to 11 species, were obtained and analyzed by the new approach and classified by sparse partial least-squares regression (PLSR). For the classification, chemical and physical information were separated by extended multiplicative signal correction and used together to build a classification model. A training set of 260 spectra and an independent test set of 130 spectra were used. Robust sparse classification models allowing the biochemical interpretation of the classification were obtained by the sparse PLSR, because only a subset of variables was retained for the analysis. With accuracy values of 95% and 98%, for the independent test set and full cross-validation respectively, the method is outperforming the previously published studies on development of an automated pollen analysis. Since the method is compatible with standard air-samplers, it can be employed with minimal modification in regular aerobiology studies. When compared with optical microscopy, which is the benchmark method in pollen analysis, the infrared microspectroscopy method offers better taxonomic resolution, as well as faster, more economical, and bias-free measurement.

Detecting intention to walk in stroke patients from pre-movement EEG correlates.

BackgroundMost studies in the field of brain-computer interfacing (BCI) for lower limbs rehabilitation are carried out with healthy subjects, even though insights gained from healthy populations may not generalize to patients in need of a BCI.MethodsWe investigate the ability of a BCI to detect the intention to walk in stroke patients from pre-movement EEG correlates. Moreover, we also investigated how the motivation of the patients to execute a task related to the rehabilitation therapy affects the BCI accuracy. Nine chronic stroke patients performed a self-initiated walking task during three sessions, with an intersession interval of one week.ResultsUsing a decoder that combines temporal and spectral sparse classifiers we detected pre-movement state with an accuracy of 64 % in a range between 18 % and 85.2 %, with the chance level at 4 %. Furthermore, we found a significantly strong positive correlation (r = 0.561, p = 0.048) between the motivation of the patients to perform the rehabilitation related task and the accuracy of the BCI detector of their intention to walk.ConclusionsWe show that a detector based on temporal and spectral features can be used to classify pre-movement state in stroke patients. Additionally, we found that patients’ motivation to perform the task showed a strong correlation to the attained detection rate of their walking intention.

Weighted-Fusion-Based Representation Classifiers for Hyperspectral Imagery

Spatial texture features have been demonstrated to be very useful for the recently-proposed representation-based classifiers, such as the sparse representation-based classifier (SRC) and nearest regularized subspace (NRS). In this work, a weighted residual-fusion-based strategy with multiple features is proposed for these classifiers. Multiple features include local binary patterns (LBP), Gabor features, and the original spectral signatures. In the proposed classification framework, representation residuals for a testing pixel from using each type of features are weighted to generate the final representation residual, and then the label of the testing pixel is determined according to the class yielding the minimum final residual. The motivation of this work is that different features represent pixels from different perspectives and their fusion in the residual domain can enhance the discriminative ability. Experimental results of several real hyperspectral image datasets demonstrate that the proposed residual-based fusion outperforms the original NRS, SRC, support vector machine (SVM) with LBP, and SVM with Gabor features, even in small-sample-size (SSS) situations.

Angular Discriminant Analysis for Hyperspectral Image Classification

Hyperspectral imagery consists of hundreds or thousands of densely sampled spectral bands. The resulting spectral information can provide unique spectral “signatures” of different materials present in a scene, which makes hyperspectral imagery especially suitable for classification problems. To fully and effectively exploit discriminative information in such images, dimensionality reduction is typically undertaken as a preprocessing before classification. Different from traditional dimensionality reduction methods, we propose angular discriminant analysis (ADA), which seeks to find a subspace that best separates classes in an angular sense—specifically, one that minimizes the ratio of between-class inner product to within-class inner product of data samples on a unit hypersphere in the resulting subspace. In this paper, we also propose local angular discriminant analysis (LADA), which preserves the locality of data in the projected space through an affinity matrix, while angularly separating different class samples. ADA and LADA are particularly useful for classifiers that rely on angular distance, such as the cosine angle distance based nearest neighbor-based classifier and sparse representation-based classifier, in which the sparse representation coefficients are learned via orthogonal matching pursuit. We also show that ADA and LADA can be easily extended to their kernelized variants by invoking the kernel trick. Experimental results based on benchmarking hyperspectral datasets show that our proposed methods are greatly beneficial as a dimensionality reduction preprocessing to the popular classifiers.

Classification of LV wall motion in cardiac MRI using kernel Dictionary Learning with a parametric approach.

In this paper, we propose a parametric approach for the assessment of wall motion in Left Ventricle (LV) function in cardiac cine-Magnetic Resonance Imaging (MRI). Time-signal intensity curves (TSICs) are identified in Spatio-temporal image profiles extracted from different anatomical segments in a cardiac MRI sequence. Different parameters are constructed from specific TSICs that present a decreasing then increasing shape reflecting dynamic information of the LV contraction. The parameters extracted from these curves are related to: 1) an average curve based on a clustering process, 2) curve skewness and 3) cross correlation values between each average clustered curve and a patient-specific reference. Several tests are performed in order to construct different vectors to train a sparse classifier based on kernel Dictionary Learning (DL). Results are compared with other classifiers like Support Vector Machine (SVM) and Discriminative Dictionary Learning. The best classification performance is obtained with information of skewness and the average curve with an accuracy about 94% using the mentioned sparse based kernel DL with a radial basis function kernel.

Practical comparison of sparse methods for classification of Arabica and Robusta coffee species using near infrared hyperspectral imaging

In the present work sparse-based methods are applied to the analysis of hyperspectral images with the aim at studying their capability of being adequate methods for variable selection in a classification framework. The key aspect of sparse methods is the possibility of performing variable selection by forcing the model coefficients related to irrelevant variables to zero. In particular, two different sparse classification approaches, i.e. sPCA+kNN and sPLS-DA, were compared with the corresponding classical methods (PCA+kNN and PLS-DA) to classify Arabica and Robusta coffee species. Green coffee samples were analyzed using near infrared hyperspectral imaging and the average spectra from each hyperspectral image were used to build training and test sets; furthermore a test image was used to evaluate the performances of the considered methods at pixel-level. In our case, sparse methods led to similar results as classical methods, with the advantage of obtaining more interpretable and parsimonious models. An important result to highlight is that variable selection performed with two different sparse classification approaches converged to the selection of same spectral regions, which implies the chemical relevance of those regions in the discrimination of Arabica and Robusta coffee species.

Continuous detection of the self-initiated walking pre-movement state from EEG correlates without session-to-session recalibration

Objective. Brain–computer interfaces (BCI) as a rehabilitation tool have been used to restore functions in patients with motor impairments by actively involving the central nervous system and triggering prosthetic devices according to the detected pre-movement state. However, since EEG signals are highly variable between subjects and recording sessions, typically a BCI is calibrated at the beginning of each session. This process is inconvenient especially for patients suffering locomotor disabilities in maintaining a bipedal position for a longer time. This paper presents a continuous EEG decoder of a pre-movement state in self-initiated walking and the usage of this decoder from session to session without recalibrating. Approach. Ten healthy subjects performed a self-initiated walking task during three sessions, with an intersession interval of one week. The implementation of our continuous decoder is based on the combination of movement-related cortical potential (MRCP) and event-related desynchronization (ERD) features with sparse classification models. Main results. During intrasession our technique detects the pre-movement state with 70% accuracy. Moreover this decoder can be applied from session to session without recalibration, with a decrease in performance of about 4% on a one- or two-week intersession interval. Significance. Our detection model operates in a continuous manner, which makes it a straightforward asset for rehabilitation scenarios. By using both temporal and spectral information we attained higher detection rates than the ones obtained with the MRCP and ERD detection models, both during the intrasession and intersession conditions.

Sparse classification of rotating machinery faults based on compressive sensing strategy

Rotating machinery is integral to production processes and is composed of several components susceptible to malfunction. Safety and stable operation necessitate early identification of potential faults. Fault identification and classifications for rotating machinery are investigated in this study utilizing expanded monitoring data. A representation classification strategy for rotating machinery faults is developed based on a newly developed compressive sensing theory focusing on extraction and classification of fault features through sparse representation combined with random dimensionality reduction mapping. Original characteristics of a vibration signal are sampled and preserved by applying a small number of random projections and a learning redundant dictionary then constructed to sparsely represent the vibration signal. Fault signal impulse information is determined through an optimization strategy with sparsity promoting. A rolling bearing is utilized as an example in the study with simulations and experiments indicating that fault characteristics may be directly extracted from a small number of random projections without ever reconstructing the vibration signal completely and with minimal prior fault characteristic frequency knowledge required.

Sparse logistic regression and polynomial modelling for detection of artificial drainage networks

Mire ditching changes dramatically mire biodiversity. Thus, drainage network detection is an important factor when analysing the natural state of a mire. In this article, we propose a method for automated drainage network detection from raster digital terrain model created from high-resolution laser scanning data. Sparse logistic regression classifier with a large generic feature set and automated feature selection is used for classification. Broken segments are connected with polynomial modelling. The results showed that our method can accurately detect artificial drainage networks.

Performance Evaluation of Different Feature Extractors and Classifiers for Recognition of Human Faces with Low Resolution Images

Face recognition is an effective biometric identification technique used in many applications such as law enforcement, document validation and video surveillance. In this paper the effect of low resolution images which are captured in real world applications, on the performance of different feature extraction techniques combined with a variety of classification approaches is evaluated. Gabor features and its combination with local phase quantization histogram (GLPQH) are dimensionality reduced by principal component analysis (PCA), linear discriminant analysis (LDA), locally sensitive discriminant analysis (LSDA) and neighbourhood preserving embedding (NPE) to extract discriminant image characteristics and the class label is attributed using the extreme learning machine (ELM), sparse classifier (SC), fuzzy nearest neighbour (FNN) or regularized discriminant classifier (RDC). ORL and AR databases are utilized and the results show that ELM and RDC have better performance and stability against resolution reduction, especially on Gabor-PCA and Gabor-LDA techniques. Among the interpolation approaches that we employed to enhance the image resolution, nearest neighbour outperforms other methods.

Multi-class classification using a signomial function

We propose two multi-class classification methods using a signomial function. Each of these methods directly constructs a multi-class classifier by solving a single optimization problem. Since the number of possible signomial terms is extremely large, we propose a column generation method that iteratively generates good signomial terms. Both of these methods obtain better or comparable classification accuracies than existing methods and also provide more sparse classifiers.

Efficient and sparse feature selection for biomedical text classification via the elastic net: Application to ICU risk stratification from nursing notes

Background and significanceSparsity is often a desirable property of statistical models, and various feature selection methods exist so as to yield sparser and interpretable models. However, their application to biomedical text classification, particularly to mortality risk stratification among intensive care unit (ICU) patients, has not been thoroughly studied. ObjectiveTo develop and characterize sparse classifiers based on the free text of nursing notes in order to predict ICU mortality risk and to discover text features most strongly associated with mortality. MethodsWe selected nursing notes from the first 24h of ICU admission for 25,826 adult ICU patients from the MIMIC-II database. We then developed a pair of stochastic gradient descent-based classifiers with elastic-net regularization. We also studied the performance-sparsity tradeoffs of both classifiers as their regularization parameters were varied. ResultsThe best-performing classifier achieved a 10-fold cross-validated AUC of 0.897 under the log loss function and full L2 regularization, while full L1 regularization used just 0.00025% of candidate input features and resulted in an AUC of 0.889. Using the log loss (range of AUCs 0.889–0.897) yielded better performance compared to the hinge loss (0.850–0.876), but the latter yielded even sparser models. DiscussionMost features selected by both classifiers appear clinically relevant and correspond to predictors already present in existing ICU mortality models. The sparser classifiers were also able to discover a number of informative – albeit nonclinical – features. ConclusionThe elastic-net-regularized classifiers perform reasonably well and are capable of reducing the number of features required by over a thousandfold, with only a modest impact on performance.

High Density Crowd Behaviors Recognition Based on Micro-behavioral and Sparse Representation

This paper presents a novel method to recognize high density crowd behaviors using micro-behaviors combining with Sparse Representation based on Locally Linear Embedding (named LLE-based Sparse Representation or LLE-SR). We extract micro-behaviors from each frame, respectively named Fountainhead, Bottleneck, Blocking, Lane and Ring/Arch, and construct micro-behaviors histograms to better describe complex high density crowd scenes. As the mid-lever semantic, micro-behaviors solve the gap between the high-level semantics and low-level semantic, and the creation of them does not need any information of target track. LLE-SR method fully considers the behavior of local manifold structure of samples. Through adding LLE regularization term in sparse classification model, the unstable property of the manifold structure can be solved, and then recognition rate is improved. Numerous experiments have been conducted in real scenes, the results of which demonstrate the effectiveness and robustness of the proposed method for high density crowd behavior recognition.