Robust Classification Approach Research Articles

A Robust Approach for Segmentation and Classification of Lung Cancer using Marker Controlled Watershed Method and Deep Hybrid Learning

A Robust Approach for Segmentation and Classification of Lung Cancer using Marker Controlled Watershed Method and Deep Hybrid Learning

Attention‐based convolutional neural network deep learning approach for robust malware classification

AbstractRecently, transforming windows files into images and its analysis using machine learning and deep learning have been considered as a state‐of‐the art works for malware detection and classification. This is mainly due to the fact that image‐based malware detection and classification is platform independent, and the recent surge of success of deep learning model performance in image classification. Literature survey shows that convolutional neural network (CNN) deep learning methods are successfully employed for image‐based windows malware classification. However, the malwares were embedded in a tiny portion in the overall image representation. Identifying and locating these affected tiny portions is important to achieve a good malware classification accuracy. In this work, a multi‐headed attention based approach is integrated to a CNN to locate and identify the tiny infected regions in the overall image. A detailed investigation and analysis of the proposed method was done on a malware image dataset. The performance of the proposed multi‐headed attention‐based CNN approach was compared with various non‐attention‐CNN‐based approaches on various data splits of training and testing malware image benchmark dataset. In all the data‐splits, the attention‐based CNN method outperformed non‐attention‐based CNN methods while ensuring computational efficiency. Most importantly, most of the methods show consistent performance on all the data splits of training and testing and that illuminates multi‐headed attention with CNN model's generalizability to perform on the diverse datasets. With less number of trainable parameters, the proposed method has achieved an accuracy of 99% to classify the 25 malware families and performed better than the existing non‐attention based methods. The proposed method can be applied on any operating system and it has the capability to detect packed malware, metamorphic malware, obfuscated malware, malware family variants, and polymorphic malware. In addition, the proposed method is malware file agnostic and avoids usual methods such as disassembly, de‐compiling, de‐obfuscation, or execution of the malware binary in a virtual environment in detecting malware and classifying malware into their malware family.

Fine-Scale Mapping of Natural Ecological Communities Using Machine Learning Approaches

Remote sensing technology has been used widely in mapping forest and wetland communities, primarily with moderate spatial resolution imagery and traditional classification techniques. The success of these mapping efforts varies widely. The natural communities of the Laurentian Mixed Forest are an important component of Upper Great Lakes ecosystems. Mapping and monitoring these communities using high spatial resolution imagery benefits resource management, conservation and restoration efforts. This study developed a robust classification approach to delineate natural habitat communities utilizing multispectral high-resolution (60 cm) National Agriculture Imagery Program (NAIP) imagery data. For accurate training set delineation, NAIP imagery, soils data and spectral enhancement techniques such as principal component analysis (PCA) and independent component analysis (ICA) were integrated. The study evaluated the importance of biogeophysical parameters such as topography, soil characteristics and gray level co-occurrence matrix (GLCM) textures, together with the normalized difference vegetation index (NDVI) and NAIP water index (WINAIP) spectral indices, using the joint mutual information maximization (JMIM) feature selection method and various machine learning algorithms (MLAs) to accurately map the natural habitat communities. Individual habitat community classification user’s accuracies (UA) ranged from 60 to 100%. An overall accuracy (OA) of 79.45% (kappa coefficient (k): 0.75) with random forest (RF) and an OA of 75.85% (k: 0.70) with support vector machine (SVM) were achieved. The analysis showed that the use of the biogeophysical ancillary data layers was critical to improve interclass separation and classification accuracy. Utilizing widely available free high-resolution NAIP imagery coupled with an integrated classification approach using MLAs, fine-scale natural habitat communities were successfully delineated in a spatially and spectrally complex Laurentian Mixed Forest environment.

A robust computational approach for jaw movement detection and classification in grazing cattle using acoustic signals

Monitoring behaviour of the grazing livestock is a difficult task because of its demanding requirements (continuous operation, large amount of information, computational efficiency, device portability, precision and accuracy) under harsh environmental conditions. Detection and classification of jaw movements (JM) events are essential for estimating information related with foraging behaviour. Acoustic monitoring is the best way to classify and quantify ruminant events related with its foraging behaviour. Although existing acoustic methods are computationally efficient, a common failure for broad applications is the deal with interference associated with environmental noises. In this work, the acoustic method, called Chew-Bite Energy Based Algorithm (CBEBA), is proposed to automatically detect and classify masticatory events of grazing cattle. The system incorporates computations of instantaneous power signal for JM-events classification associated with chews, bites and composite chew-bites, and additionally between two classes of chew events: i) low energy chews that are associated with rumination and ii) high energy chews that are associated with grazing. The results demonstrate that CBEBA achieve a recognition rate of 91.9% and 91.6% in noiseless and noisy conditions, respectively, with a high classification precision and a marginal increment of computational cost compared to previous algorithms, suggesting feasibility for implementation in low-cost embedded systems.

Joint nuclear- and [formula omitted]-norm regularized heterogeneous tensor decomposition for robust classification

Traditional approaches that represent observations in vector or matrix notation easily lead to structure information losses and dependency destruction between elements. Tensors are the generalizations of matrices, providing a natural representation of rich structure in real-world multiway arrays. Although tensor decomposition has been verified and used in classification tasks, the heterogeneity gap among different orders is often neglected. In this paper, we exploited tensor decomposition framework and presented a heterogeneous tensor decomposition for robust classification (HTDRC) approach, which integrates nuclear and ℓ2,1-norm for intrinsic representation learning. Specifically, to obtain the lowest-rank intrinsic representations of tensorial data from an underlying low-dimensional subspace, HTDRC simultaneously learns a set of orthogonality constrained factor matrices and a low-rank constrained representation matrix. To better guide the decomposition process, a robust discriminant feature selection scheme is utilized by imposing ℓ2,1-norm penalty on the constructed classification loss and the regularized terms simultaneously. Experiments on four datasets demonstrated the superior performance of our proposed approach.

A simple and robust wetland classification approach by using optical indices, unsupervised and supervised machine learning algorithms

Wetlands are important for their peat reservoir, dynamic land cover and natural resources, ecological and hydrological regimes, fossil fuels reservoir, and crucial carbon storage. The global wetlands are decreasing since 1800 due to climatic phenomena and human activities. Wetland mapping with satellite data is not new but an ongoing challenge due to its precision relies on data quality and data classification schemes. The accuracy of such mapping is emerging due to the gradual establishment of satellite data and subsequent data modeling technologies, i.e., big data modeling with machine learning (ML) algorithms. Our study introduced a simple, scalable, and robust wetland classification by applying unsupervised (K-means cluster – KMC) and supervised (Support vector machine classification – SVMc) ML algorithms. We used Landsat optical data to model normalized difference vegetation index (NDVI) and normalized difference water index (NDWI), as the primary inputs for KMC. Later KMC data were supervised by SVMc with training data from 20 field observations. The accuracy tests insights that both optical indices have considerably less error and all SVMc models have about 99% accuracy. The 1 to 1 validation insights that our wetland classification presented more detail wetland cover areas than reference data. The test case SVMc showed that Class 1 and 2 are optimally fitting, Class 3 and 4 are overfitting, and Class 5 is underfitting. Furthermore, the sensitivity analysis insight that all SVMc models are optimally fitting and SVMc is more sensitive to NDVI than NDWI. From SVMc models we can see that Selisoo bog in Estonia lost a considerable amount of wetland covers including water bodies, and more large forest covers are taking wetland areas, i.e., mixed forest and coniferous trees. Our methodological approach insights a simple and robust wetland classification based on advanced unsupervised and supervised ML algorithms despite some unavoidable limitation.

DCPM: an effective and robust approach for diabetes classification and prediction

Diabetes is the most common medical disorders that occur due to the malfunctioning of the pancreas. It increases the level of sugar in the body and poses a severe concern to human health by adversely affecting almost all major organs of the body, including kidney, heart, eyes, etc. The number of research works in the literature proves that machine learning techniques can increase the early detection of disease and decrease medical error rates to save human life. Developing an accurate and effective diabetes prediction model is always a challenge, as the medical dataset suffers from outliers and missing values. The aim of this study is to build an accurate and robust Diabetes Classification and Prediction Model (DCPM) on a dataset that suffers from the class imbalance problem and contains outliers and missing values. The proposed work devises an effective pre-processing technique to remove outliers, fill missing values, standardize data and select relevant features for model learning in a pipelined manner. The proposed pre-processing techniques were applied on the Pima Indian Diabetes (PID) dataset obtained from the University of California at Irvine (UCI) Repository. The K-NN classifier is optimized to find the optimum value of k and is then trained and evaluated on the most predictive set of features of the pre-processed PID dataset. The performance of the proposed model is assessed using classification accuracy, precision, recall and F1-score. The proposed approach is able to attain statistically good classification accuracy, recall, precision and F1-score as 92.28%, 92.36%, 92.38% and 92.31%, respectively. The proposed model outperforms existing state-of-the-art approaches in terms of accuracy. Therefore, the proposed DCPM can assist the medical experts by providing a quick, precise and reliable recommendation that can be considered while making a crucial decision about the health of a patient in the healthcare sector.

To Do or Not To Do Business with a Country: A Robust Classification Approach

In the face of global uncertainty and a growing reliance on third party indices to gain a snapshot of a country's operations, accurate decision making makes or breaks relationships in global trade. Under this aegis, we question the validity of traditional logistic regression using the maximum likelihood estimator (MLE) in classifying countries for doing business. This paper proposes that a weighted version of the Bianco and Yohai (BY) es- timator is a superlative and robust (outlier resistant) tool in the hands of practitioners to gauge the correct antecedents of a country's internal envi- ronment and decide whether to do or not do business with that country. In addition, this robust process is eective in dierentiating between \problem countries and \safe countries for doing business. An existing \R program for the BY estimation technique by Croux and Haesbroeck has been modied to t our cause.

An Unsupervised Prediction Model for Salmonella Detection with Hyperspectral Microscopy: A Multi-Year Validation

Hyperspectral microscope images (HMIs) have been previously explored as a tool for the early and rapid detection of common foodborne pathogenic bacteria. A robust unsupervised classification approach to differentiate bacterial species with the potential for single cell sensitivity is needed for real-world application, in order to confirm the identity of pathogenic bacteria isolated from a food product. Here, a one-class soft independent modelling of class analogy (SIMCA) was used to determine if individual cells are Salmonella positive or negative. The model was constructed and validated with a spectral library built over five years, containing 13 Salmonella serotypes and 14 non-Salmonella foodborne pathogens. An image processing method designed to take less than one minute paired with the one-class Salmonella prediction algorithm resulted in an overall classification accuracy of 95.4%, with a Salmonella sensitivity of 0.97, and specificity of 0.92. SIMCA’s prediction accuracy was only achieved after a robust model incorporating multiple serotypes was established. These results demonstrate the potential for HMI as a sensitive and unsupervised presumptive screening method, moving towards the early (<8 h) and rapid (<1 h) identification of Salmonella from food matrices.

A Robust and Novel Approach for Brain Tumor Classification Using Convolutional Neural Network.

Brain tumors are the most common and aggressive illness, with a relatively short life expectancy in their most severe form. Thus, treatment planning is an important step in improving patients' quality of life. In general, image methods such as computed tomography (CT), magnetic resonance imaging (MRI), and ultrasound images are used to assess tumors in the brain, lung, liver, breast, prostate, and so on. X-ray images, in particular, are utilized in this study to diagnose brain tumors. This paper describes the investigation of the convolutional neural network (CNN) to identify brain tumors from X-ray images. It expedites and increases the reliability of the treatment. Because there has been a significant amount of study in this field, the presented model focuses on boosting accuracy while using a transfer learning strategy. Python and Google Colab were utilized to perform this investigation. Deep feature extraction was accomplished with the help of pretrained deep CNN models, VGG19, InceptionV3, and MobileNetV2. The classification accuracy is used to assess the performance of this paper. MobileNetV2 had the accuracy of 92%, InceptionV3 had the accuracy of 91%, and VGG19 had the accuracy of 88%. MobileNetV2 has offered the highest level of accuracy among these networks. These precisions aid in the early identification of tumors before they produce physical adverse effects such as paralysis and other impairments.

A Comparison of Gaofen-2 and Sentinel-2 Imagery for Mapping Mangrove Forests Using Object-Oriented Analysis and Random Forest

Mangrove forest (MF) extents and distributions are fundamental for conservation and restoration efforts. According to previous studies, both the commercial Gaofen-2 (GF-2) imagery (0.8 m spatial resolution and 4 spectral bands) and freely accessed Sentinel-2 (S2) imagery (10 m spatial resolution and 13 spectral bands) have been successfully used to map MFs. However, the efficiency and accuracy of MF mapping based on these two data is not clear, especially for large-scale applications. To address this issue, first, we developed a robust classification approach by integrating object-based image analysis (OBIA) and random forest (RF) algorithm; and then, applied this approach to GF-2 and S2 images to map the extents of MF along the entire coasts of Guangxi, China, respectively; at last, compared the efficiency and accuracy of GF-2 and S2 imagery in MF mapping. Results showed that: first, based on OBIA and RF integrated classification approach both MF maps derived from GF-2 and S2 obtained high mapping accuracies (the overall accuracy was 96% and 94%, respectively); second, areal extent of MFs in Guangxi extracted from GF-2 and S2 images was 8182 and 8040 ha, respectively; third, GF-2 imagery has extraordinary abilities in detecting fragmented MF patches located along landward and seaward edges; and finally, S2 imagery performed better in detecting seaward submerged MFs and separating MF from terrestrial vegetation. Results and conclusions of this study can provide basic considerations for selecting appropriate data source in MF or wetland vegetation mapping tasks.

FaVAD: A software workflow for characterization and visualizing of defects in crystalline structures

The analysis of defects and defect dynamics in crystalline materials is important for fundamental science and for a wide range of applied engineering. With increasing system size the analysis of molecular-dynamics simulation data becomes non-trivial. Here, we present a workflow for semi-automatic identification and classification of defects in crystalline structures, combining a new approach for defect description with several already existing open-source software packages. Our approach addresses the key challenges posed by the often relatively tiny volume fraction of the modified parts of the sample, thermal motion and the presence of potentially unforeseen atomic configurations (defect types) after irradiation. The local environment of any atom is converted into a rotation-invariant descriptive vector (‘fingerprint’), which can be compared to known defect types and also yields a distance metric suited for classification. Vectors which cannot be associated to known structures indicate new types of defects. As proof-of-concept we apply our method on an iron sample to analyze the defects caused by a collision cascade induced by a 10 keV primary-knock-on-atom. The obtained results are in good agreement with reported literature values. Program summaryProgram Title: Fingerprinting and Visualization Analyzer of Defects (FaVAD).CPC Library link to program files:https://doi.org/10.17632/bmv9kxkzg3.1Developer’s repository link:http://gitlab.mpcdf.mpg.de/NMPP/favad.gitLicensing provisions: GPLv3.Programming language: Python 3, Fortran 90, and C++.Nature of problem: The analysis of damage and damage evolution in crystalline materials is important for fundamental science and for a wide range of applied engineering. Defects in materials on an atomic level are commonly analyzed by Wigner–Seitz or topology Voronoi tessellation based methods. However, these approaches exhibit specific shortcoming, especially at elevated sample temperatures. In order to improve upon that, a more robust and quantifiable identification and classification approach of known as well as of unpredicted defect structures is desirable.Solution method: A fingerprint-like method is proposed to analyze in detail the damage in a material augmented with a probabilistic interpretation. It is based on the calculation of a descriptor vector for each atom in the sample. These vectors represent in a compact form the individual environments of the atoms.For standard types of defects (i.e. interstitial atoms) the corresponding descriptor vectors can be precomputed and used for rapid classification. Unexpected or less common defect types can be identified by applying a principal component analysis to the descriptor vectors. Vacancies in the material are identified by computing the radii of the largest empty spheres which can be embedded into the sample, followed by a thresholding process. This new method is easy to use and requires only modest computational resources.Finally, the classified defects are visualized using the open source software VisIt.Additional comments including restrictions and unusual features: The descriptor vectors are computed using the command line interface of QUIP with the Gaussian Approximation potential (GAP) package. The analysis of the sample is done using Python scripts which make extensive use of the NumPy package. A modified KDTree2 code is employed to calculate the location of single vacancies and voids. The program VisIt is used for the visualization of the classified point defects in the sample. We provide a Dockerfile for automatically creating a portable Docker container which installs all the programs together with a Python script to analyze as an example a damaged iron molecular dynamics sample. A shell script to install the programs locally in a Linux-based server or desktop environment is included also.

Multi-geophysical parameter classification of the Montserrat geothermal system

Multi-geophysical parameter classification can help to reduce the uncertainties of interpretations that often rely on one geophysical technique. Integrating these varying datasets requires a more robust classification approach rather than traditional qualitative methods. In this study, we applied the Fuzzy c-means (FCM) method to quantitatively classify similarities in a high resolution seismic tomography, a magnetotellurics and gravity datasets obtained in Montserrat. To group similar datapoints, this application uses a Euclidean distance measure and a membership function. Assigned membership values indicate the degree to which a datapoint belongs to a specific class. The spatial distribution of the derived classes, each classified with distinct geophysical parameters, helped to provide new structural and petrological information of the Montserrat geothermal system. In comparison to previous models, our new cluster model highlights two major improvements. These include the resolution and assessment of the spatial extension and 3D geometry of previously undetected features within the Montserrat geothermal system and the constrain and characterization of earlier identified anomalies. We additionally utilized geological and petrological data obtained from three geothermal wells in the Montserrat geothermal system to help validate our classifications. Based on a semi-quantitative approach we assessed the reliability of the FCM technique in relation to the likely uncertainties of the different geophysical models.

A robust deep learning approach for automatic classification of seizures against non-seizures

Identifying epileptic seizures through analysis of the electroencephalography (EEG) signal becomes a standard method for the diagnosis of epilepsy. Manual seizure identification on EEG by trained neurologists is time-consuming, labor-intensive and a reliable automatic seizure/non-seizure classification method is needed. One of the challenges in automatic seizure/non-seizure classification is that seizure morphologies exhibit considerable variabilities. In order to capture essential seizure patterns, this paper leverages an attention mechanism and a bidirectional long short-term memory (BiLSTM) to exploit both spatial and temporal discriminating features and overcome seizure variabilities. The attention mechanism captures spatial features according to the contributions of different brain regions to seizures. BiLSTM extracts discriminating temporal features in forward and backward directions. Cross-validation experiments and cross-patient experiments over the noisy data of CHB-MIT were performed. We obtained average sensitivity of 87.30%, specificity of 88.30% and precision of 88.29% in cross-validation experiments, higher than using the current state-of-the-art methods, and the standard deviations were lower. These results indicate that our approach performs well against current state-of-the-art methods and is more robust across patients.

H2K – A robust and optimum approach for detection and classification of groundnut leaf diseases

One of the greatest key factors contributing to low yield and destruction of groundnut plant growth is leaf disease attack. The groundnut plants are prone to the diseases such as fungi, soil borne and viruses. The leaf disease detection and identification in the early stage is needed to control the spread of infection and also helps to get the maximum yield. The detection and classification of groundnut leaf diseases through naked eye observation by an expert is expensive that too developing countries. So, providing a software based solution for the above task is of great significance. In this paper, an image processing based approach that automatically identifies and categorizes the groundnut leaf diseases has been presented. The proposed method H2K is a fusion of Harris corner detector, HOG (Histogram on Oriented Gradient) and KNN classifier for accurate detection and classification of groundnut leaf diseases. It comprises of number of steps viz. image acquisition, image pre-processing by applying binary mask, HSV segmentation to segment the disease affected part, features detection and extraction using H2K (Harris, HOG and K-Nearest Neighbor) based classification of groundnut leaf diseases. The existing works concentrates on leaf diseases that are commonly occurring in any crops, but this paper proposed a robust and optimum approach for detection and classification of all major leaf diseases for Groundnut crop which is first of its kind. Thus the H2K method aids to improve the crop production and maximizes the yield. The proposed method H2K works well on sample images and detects and classifies 5 major groundnut leaf diseases including late spot which is difficult to control if it is not identified in the early stages. The existing Multiclass SVM is taken for comparison and the examined results shows that the H2K is robust and optimum in classifying groundnut leaf diseases with improved accuracy of 97.67%.

EnClaSC: a novel ensemble approach for accurate and robust cell-type classification of single-cell transcriptomes

BackgroundIn recent years, the rapid development of single-cell RNA-sequencing (scRNA-seq) techniques enables the quantitative characterization of cell types at a single-cell resolution. With the explosive growth of the number of cells profiled in individual scRNA-seq experiments, there is a demand for novel computational methods for classifying newly-generated scRNA-seq data onto annotated labels. Although several methods have recently been proposed for the cell-type classification of single-cell transcriptomic data, such limitations as inadequate accuracy, inferior robustness, and low stability greatly limit their wide applications.ResultsWe propose a novel ensemble approach, named EnClaSC, for accurate and robust cell-type classification of single-cell transcriptomic data. Through comprehensive validation experiments, we demonstrate that EnClaSC can not only be applied to the self-projection within a specific dataset and the cell-type classification across different datasets, but also scale up well to various data dimensionality and different data sparsity. We further illustrate the ability of EnClaSC to effectively make cross-species classification, which may shed light on the studies in correlation of different species. EnClaSC is freely available at https://github.com/xy-chen16/EnClaSC.ConclusionsEnClaSC enables highly accurate and robust cell-type classification of single-cell transcriptomic data via an ensemble learning method. We expect to see wide applications of our method to not only transcriptome studies, but also the classification of more general data.

Dual subspace discriminative projection learning

In this paper, we propose a dual subspace discriminative projection learning (DSDPL) framework for multi-category image classification. Our approach reflects the notion that images are composed of class-shared information, class-specific information, and sparse noise. Unlike traditional subspace learning methods, DSDPL serves to decompose original high dimensional data, via learned projection matrices, into class-shared and class-specific subspaces. The learned projection matrices are jointly constrained with l2,1 sparse norm and LDA terms while the reconstructive properties of DSDPL reduce information loss, leading to greater stability within low dimensional subspaces. Regression-based terms are also included to facilitate a more robust classification approach, using extracted class-specific features for better classification. Our approach is examined on five different datasets for face, object and scene classifications. Experimental results demonstrate not only the superiority and versatility of DSDPL over current benchmark approaches, but also a more robust classification approach with low sample size training data.

Tracking long-term floodplain wetland changes: A case study in the China side of the Amur River Basin

Floodplain wetlands in the China side of the Amur River Basin (CARB) undergone consistent decreases because of both natural and anthropogenic drivers. Monitoring floodplain wetlands dynamics and conversions over long-time periods is thus fundamental to sustainable management and protection. Due to complexity and heterogeneity of floodplain environments, however, it is difficult to map wetlands accurately over a large area as the CARB. To address this issue, we developed a novel and robust classification approach integrating image compositing algorithm, objected-based image analysis, and hierarchical random forest classification, named COHRF, to delineate floodplain wetlands and surrounding land covers. Based on the COHRF classification approach, 4622 Landsat images were applied to produce a 30-m resolution dataset characterizing dynamics and conversions of floodplain wetlands in the CARB during 1990–2018. Results show that (1) all floodplain land cover maps in 1990, 2000, 2010, and 2018 had high mapping accuracies (ranging from 90 %±0.001–97%±0.005), suggesting that COHRF is a robust classification approach; (2) CARB experienced an approximately 25 % net loss of floodplain wetlands with an area declined from 8867 km2 to 6630 km2 during 1990–2018; (3) the lost floodplain wetlands were mostly converted into croplands, while, there were 111 km2 and 256 km2 of wetlands rehabilitated from croplands during periods of 2000–2010 and 2010–2018, respectively. To our knowledge, this study is the first attempt that focus on delineating floodplain wetlands at a large-scale and produce the first 30-m spatial resolution dataset demonstrating long-term dynamics of floodplain wetlands in the CARB. The COHRF classification approach could be used to classify other ecosystems readily and robustly. The resultant dataset will contribute to sustainable use and conservation of wetlands in the Amur River Basin and provide essential information for related researches.

Classification of regional and genre varieties of Chinese: A correspondence analysis approach based on comparable balanced corpora

AbstractThis paper proposes a robust text classification and correspondence analysis approach to identification of similar languages. In particular, we propose to use the readily available information of clauses and word length distribution to model similar languages. The modeling and classification are based on the hypothesis that languages are self-adaptive complex systems and hence can be classified by dynamic features describing the system, especially in terms of distributional relations of constituents of a system. For similar languages whose grammatical differences are often subtle, classification based on dynamic system features should be more effective. To test this hypothesis, we considered both regional and genre varieties of Mandarin Chinese for classification. The data are extracted from two comparable balanced corpora to minimize possible confounding factors. The two corpora are the Sinica Corpus from Taiwan and the Lancaster Corpus of Mandarin Chinese from Mainland China, and the two genres are reportage and review. Our text classification and correspondence analysis results show that the linguistically felicitous two-level constituency model combining power functions between word and clauses effectively classifies the two varieties of Chinese for both genres. In addition, we found that genres do have compounding effect on classification of regional varieties. In particular, reportage in two varieties is more likely to be classified than review, corroborating the complex system view of language variations. That is, language variations and changes typically do not take place evenly across the board for the complete language system. This further enhances our hypothesis that dynamic complex system features, such as the power functions captured by the Menzerath–Altmann law, provide effective models in classifications of similar languages.

A Robust Pneumonia Classification Approach based on Self-Paced Learning

This study proposes a self-paced learning scheme that integrates self-training and deep learning to select and learn labeled and unlabeled data samples for classifying anterior-posterior chest images as either being pneumonia-infected or normal. With this new approach, a model is first trained with labeled data. The model is evaluated on unlabeled data to generate pseudo labels for the unlabeled data. Using a novel selection scheme, the pseudo-labeled samples are then selected to update the model in next training iteration of the semi-supervised training process. The selected pseudo-labeled images to be added to the next training iteration are images with the most confident probabilities from every unlabeled class. Such a selection scheme prevents mistake reinforcement, which is a prevalent occurrence in self-training. With deep models having the tendency to latch onto well-represented class samples while ignoring less transferable and represented classes, especially in the case of unbalanced data, the proposed method utilizes a novel algorithm for the generation and selection of reliable top-K pseudo-labeled samples to be used in updating the model during the next training phase. Such an approach does not only force the model to learn the hard samples in the training data, it also helps enlarge the training set by generating enough samples that satisfy the hunger of deep models. Extensive experimental evaluation of the proposed method yields higher accuracy results compared to methods mentioned in the literature on the same dataset, an indication of the effectiveness of the proposed method.