Multi-class Classification Task Research Articles

Fisher Regularized ε-Dragging for Image Classification

Discriminative least squares regression (DLSR) has been shown to achieve promising performance in multi-class image classification tasks. Its key idea is to force the regression labels of different classes to move in opposite directions by means of the ε-dragging technique, yielding discriminative regression model exhibiting wider margins. However, the ε-dragging technique ignores an important problem: its relaxation matrix is dynamically updated in optimization, which means the dragging values can also cause the labels from the same class to be uncorrelated. In order to learn a more powerful projection, as well as regression labels, we propose a Fisher regularized ε-dragging framework (Fisher-ε) for image classification by constraining the relaxed labels using the Fisher criterion. On one hand, the Fisher criterion improves the intra-class compactness of the relaxed labels during relaxation learning. On the other hand, it is expected further to enhance the inter-class separability of ε-dragging. Fisher-ε for the first time ever attempts to integrate the Fisher criterion and ε-dragging technique into a unified model because they are complementary in learning discriminative projection. Extensive experiments on various datasets demonstrate that the proposed Fisher-ε method achieves performance that is superior to other state-of-the-art classification methods. The Matlab codes are available at https://github.com/chenzhe207/Fisher-epsilon.

Noise effects analysis on subspace-based damage detection with neural networks

Structural health monitoring (SHM) is nowadays a significant and topical issue in the civil engineering field, whose first-level task is the damage detection job. Recent advances in artificial intelligence (AI) combined with non-destructive assessments based on output-only vibration signals represent the frontier of SHM. The present research study has handled an AI-based multi-class damage classification task on a numerical benchmark beam problem with three different procedures. The main purpose was to investigate the noise effects of the analyzed damage detection procedures simulating real-world accelerometer sensors. The first damage identification method leverages simple statistical features of vibration signals, whereas the second method additionally considers a subspace-based damage indicator. The third method relies on a dataset composed of a specific set of damage indicators. Two deep learning (DL) techniques have been adopted to compare and verify the results, i.e. a multi-layer perceptron (MLP) and a one-dimensional convolutional network (1D-CNN). Three groups of different reasonable signal-to-noise ratio (SNR) levels were set up to simulate micro electro-mechanical system (MEMS) accelerometer sensors in real-world conditions to evaluate noise effects. In a nutshell, the results demonstrated that the DL models based on stochastic subspace identification (SSI) methods have good noise immunity performance for damage detection purposes.

Drug repurposing and prediction of multiple interaction types via graph embedding

BackgroundFinding drugs that can interact with a specific target to induce a desired therapeutic outcome is key deliverable in drug discovery for targeted treatment. Therefore, both identifying new drug–target links, as well as delineating the type of drug interaction, are important in drug repurposing studies.ResultsA computational drug repurposing approach was proposed to predict novel drug–target interactions (DTIs), as well as to predict the type of interaction induced. The methodology is based on mining a heterogeneous graph that integrates drug–drug and protein–protein similarity networks, together with verified drug-disease and protein-disease associations. In order to extract appropriate features, the three-layer heterogeneous graph was mapped to low dimensional vectors using node embedding principles. The DTI prediction problem was formulated as a multi-label, multi-class classification task, aiming to determine drug modes of action. DTIs were defined by concatenating pairs of drug and target vectors extracted from graph embedding, which were used as input to classification via gradient boosted trees, where a model is trained to predict the type of interaction. After validating the prediction ability of DT2Vec+, a comprehensive analysis of all unknown DTIs was conducted to predict the degree and type of interaction. Finally, the model was applied to propose potential approved drugs to target cancer-specific biomarkers.ConclusionDT2Vec+ showed promising results in predicting type of DTI, which was achieved via integrating and mapping triplet drug–target–disease association graphs into low-dimensional dense vectors. To our knowledge, this is the first approach that addresses prediction between drugs and targets across six interaction types.

MCSC-Net: COVID-19 detection using deep-Q-neural network classification with RFNN-based hybrid whale optimization.

COVID-19 is the most dangerous virus, and its accurate diagnosis saves lives and slows its spread. However, COVID-19 diagnosis takes time and requires trained professionals. Therefore, developing a deep learning (DL) model on low-radiated imaging modalities like chest X-rays (CXRs) is needed. The existing DL models failed to diagnose COVID-19 and other lung diseases accurately. This study implements a multi-class CXR segmentation and classification network (MCSC-Net) to detect COVID-19 using CXR images. Initially, a hybrid median bilateral filter (HMBF) is applied to CXR images to reduce image noise and enhance the COVID-19 infected regions. Then, a skip connection-based residual network-50 (SC-ResNet50) is used to segment (localize) COVID-19 regions. The features from CXRs are further extracted using a robust feature neural network (RFNN). Since the initial features contain joint COVID-19, normal, pneumonia bacterial, and viral properties, the conventional methods fail to separate the class of each disease-based feature. To extract the distinct features of each class, RFNN includes a disease-specific feature separate attention mechanism (DSFSAM). Furthermore, the hunting nature of the Hybrid whale optimization algorithm (HWOA) is used to select the best features in each class. Finally, the deep-Q-neural network (DQNN) classifies CXRs into multiple disease classes. The proposed MCSC-Net shows the enhanced accuracy of 99.09% for 2-class, 99.16% for 3-class, and 99.25% for 4-class classification of CXR images compared to other state-of-art approaches. The proposed MCSC-Net enables to conduct multi-class segmentation and classification tasks applying to CXR images with high accuracy. Thus, together with gold-standard clinical and laboratory tests, this new method is promising to be used in future clinical practice to evaluate patients.

Abusive Content Detection in Arabic Tweets Using Multi-Task Learning and Transformer-Based Models

Different social media platforms have become increasingly popular in the Arab world in recent years. The increasing use of social media, however, has also led to the emergence of a new challenge in the form of abusive content, including hate speech, offensive language, and abusive language. Existing research work focuses on automatic abusive content detection as a binary classification problem. In addition, the existing research work on the automatic detection task surrounding abusive Arabic content fails to tackle the dialect-specific phenomenon. Consequently, this has led to two important issues in the automatic abusive Arabic content detection task. In this study, we used a multi-aspect annotation schema to tackle the automatic abusive content detection problem in Arabic countries, based on the multi-class classification task and the dialectal Arabic (DA)-specific phenomenon. More precisely, the multi-aspect annotation schema includes five attributes: directness, hostility, target, group, and annotator. We specifically developed a framework to automatically detecting abusive content on Twitter using natural language processing (NLP) techniques. The developed framework used different models of machine learning (ML), deep learning (DL), and pretrained Arabic language models (LMs) using the multi-aspect annotation dataset. In addition, to investigate the impact of the other approaches, such as multi-task learning (MTL), we developed four MTL models built on top of a pretrained DA language model (called MARBERT) and trained on the multi-aspect annotation dataset. Our MTL models and pretrained Arabic LMs enhanced the performance compared to the existing DL model mentioned in the literature.

Deep-Stacked Convolutional Neural Networks for Brain Abnormality Classification Based on MRI Images.

An automated diagnosis system is crucial for helping radiologists identify brain abnormalities efficiently. The convolutional neural network (CNN) algorithm of deep learning has the advantage of automated feature extraction beneficial for an automated diagnosis system. However, several challenges in the CNN-based classifiers of medical images, such as a lack of labeled data and class imbalance problems, can significantly hinder the performance. Meanwhile, the expertise of multiple clinicians may be required to achieve accurate diagnoses, which can be reflected in the use of multiple algorithms. In this paper, we present Deep-Stacked CNN, a deep heterogeneous model based on stacked generalization to harness the advantages of different CNN-based classifiers. The model aims to improve robustness in the task of multi-class brain disease classification when we have no opportunity to train single CNNs on sufficient data. We propose two levels of learning processes to obtain the desired model. At the first level, different pre-trained CNNs fine-tuned via transfer learning will be selected as the base classifiers through several procedures. Each base classifier has a unique expert-like character, which provides diversity to the diagnosis outcomes. At the second level, the base classifiers are stacked together through neural network, representing the meta-learner that best combines their outputs and generates the final prediction. The proposed Deep-Stacked CNN obtained an accuracy of 99.14% when evaluated on the untouched dataset. This model shows its superiority over existing methods in the same domain. It also requires fewer parameters and computations while maintaining outstanding performance.

Human Not in the Loop: Objective Sample Difficulty Measures for Curriculum Learning.

Curriculum learning is a learning method that trains models in a meaningful order from easier to harder samples. A key here is to devise automatic and objective difficulty measures of samples. In the medical domain, previous work applied domain knowledge from human experts to qualitatively assess classification difficulty of medical images to guide curriculum learning, which requires extra annotation efforts, relies on subjective human experience, and may introduce bias. In this work, we propose a new automated curriculum learning technique using the variance of gradients (VoG) to compute an objective difficulty measure of samples and evaluated its effects on elbow fracture classification from X-ray images. Specifically, we used VoG as a metric to rank each sample in terms of the classification difficulty, where high VoG scores indicate more difficult cases for classification, to guide the curriculum training process We compared the proposed technique to a baseline (without curriculum learning), a previous method that used human annotations on classification difficulty, and anti-curriculum learning. Our experiment results showed comparable and higher performance for the binary and multi-class bone fracture classification tasks.

Differential Evolution-Based Feature Selection: A Niching-Based Multiobjective Approach

Feature selection is to reduce both the dimensionality of data and the classification error rate (i.e., increase the classification accuracy) of a learning algorithm. The two objectives are often conflicting, thus a multiobjective feature selection method can obtain a set of nondominated feature subsets. Each solution in the set has a different size and a corresponding classification error rate. However, most existing feature selection algorithms have ignored that, for a given size, there can be different feature subsets with very similar or the same accuracy. This article introduces a niching-based multiobjective feature selection method that simultaneously minimizes the number of selected features and the classification error rate. The proposed method conceives to identify: 1) a set of feature subsets with good convergence and distribution and 2) multiple feature subsets choosing the same number of features with almost the same lowest classification error rate. The contributions of this article are threefold. First, a niching and global interaction mutation operator is proposed that can produce promising feature subsets. Second, a newly developed environmental selection mechanism allows equal informative feature subsets to be stored by relaxing the Pareto-dominance relationship. Finally, the proposed subset repairing mechanism can generate better feature subsets and further remove the redundant features. The proposed method is compared against seven multiobjective feature selection algorithms on 19 datasets, including both binary and multiclass classification tasks. The results show that the proposed method can evolve a rich and diverse set of nondominated solutions for different feature selection tasks, and their availability helps in understanding the relationships between features.

Image-based features in machine learning to identify delivery errors and predict error magnitude for patient-specific IMRT quality assurance

ObjectiveTo identify delivery error type and predict associated error magnitude by image-based features using machine learning (ML).MethodsIn this study, a total of 40 thoracic plans (including 208 beams) were selected, and four error types with different magnitudes were introduced into the original plans, including 1) collimator misalignment (COLL), 2) monitor unit (MU) variation, 3) systematic multileaf collimator misalignment (MLCS), and 4) random MLC misalignment (MLCR). These dose distributions of portal dose predictions for the original plans were defined as the reference dose distributions (RDD), while those for the error-introduced plans were defined as the error-introduced dose distributions (EDD). Both distributions were calculated for all beams with portal dose image prediction (PDIP). Besides, 14 image-based features were extracted from RDD and EDD of portal dose predictions to obtain the feature vectors. In addition, a random forest was adopted for the multiclass classification task, and regression prediction for error magnitude.ResultsThe top five features extracted with the highest weight included 1) the relative displacement in the x direction, 2) the ratio of the absolute minimum residual error to the maximal RDD value, 3) the product of the maximum and minimum residuals, 4) the ratio of the absolute maximum residual error to the maximal RDD value, and 5) the ratio of the absolute mean residual value to the maximal RDD value. The relative displacement in the x direction had the highest weight. The overall accuracy of the five-class classification model was 99.85% for the validation set and 99.30% for the testing set. This model could be applied to the classification of the error-free plan, COLL, MU, MLCS, and MLCR with an accuracy of 100%, 98.4%, 99.9%, 98.0%, and 98.3%, respectively. MLCR had the worst performance in error magnitude prediction (70.1–96.6%), while others had better performance in error magnitude prediction (higher than 93%). In the error magnitude prediction, the mean absolute error (MAE) between predicted error magnitude and actual error ranged from 0.03 to 0.33, with the root mean squared error (RMSE) varying from 0.17 to 0.56 for the validation set. The MAE and RMSE ranged from 0.03 to 0.50 and 0.44 to 0.59 for the test set, respectively.ConclusionIt could be demonstrated in this study that the image-based features extracted from RDD and EDD can be employed to identify different types of delivery errors and accurately predict error magnitude with the assistance of ML techniques. They can be used to associate traditional gamma analysis with clinically based analysis for error classification and magnitude prediction in patient-specific IMRT quality assurance.

Multi-view feature representation and fusion for drug-drug interactions prediction

BackgroundDrug-drug interactions (DDIs) prediction is vital for pharmacology and clinical application to avoid adverse drug reactions on patients. It is challenging because DDIs are related to multiple factors, such as genes, drug molecular structure, diseases, biological processes, side effects, etc. It is a crucial technology for Knowledge graph to present multi-relation among entities. Recently some existing graph-based computation models have been proposed for DDIs prediction and get good performance. However, there are still some challenges in the knowledge graph representation, which can extract rich latent features from drug knowledge graph (KG).ResultsIn this work, we propose a novel multi-view feature representation and fusion (MuFRF) architecture to realize DDIs prediction. It consists of two views of feature representation and a multi-level latent feature fusion. For the feature representation from the graph view and KG view, we use graph isomorphism network to map drug molecular structures and use RotatE to implement the vector representation on bio-medical knowledge graph, respectively. We design concatenate-level and scalar-level strategies in the multi-level latent feature fusion to capture latent features from drug molecular structure information and semantic features from bio-medical KG. And the multi-head attention mechanism achieves the optimization of features on binary and multi-class classification tasks. We evaluate our proposed method based on two open datasets in the experiments. Experiments indicate that MuFRF outperforms the classic and state-of-the-art models.ConclusionsOur proposed model can fully exploit and integrate the latent feature from the drug molecular structure graph (graph view) and rich bio-medical knowledge graph (KG view). We find that a multi-view feature representation and fusion model can accurately predict DDIs. It may contribute to providing with some guidance for research and validation for discovering novel DDIs.

Semi-Supervised Learning for Medical Image Classification Based on Anti-Curriculum Learning

Although deep learning has achieved great success in image classification, large amounts of labelled data are needed to make full use of the advantages of deep learning. However, annotating a large number of images is expensive and time-consuming, especially annotating medical images, which requires professional knowledge. Therefore, semi-supervised learning has shown its potential for medical image classification. In this paper, we propose a novel pseudo-labelling semi-supervised learning method for medical image classification. Firstly, we utilize the anti-curriculum strategy for model training to prevent the model from producing predictions with a high value from the samples which are similar with existed labeled data. Secondly, to produce more stable and accurate pseudo labels for unlabeled data, we generate the pseudo labels with ensemble predictions provided by the model with samples augmented by different augmentations. In addition, we refine the generated pseudo labels using the prediction of the model at the current epoch in order to make the model learn from itself and improve the model performance. Comparative experiments on the Chest X-ray14 dataset for a multi-label classification task and the ISIC 2018 dataset for a multi-class classification task are performed, and the experimental results show the effectiveness of our method.

Domain generalization in nematode classification

Nematode images captured by different microscopes may appear differently in terms of image color and image quality, resulting in these images laying in different learning domains. This can negatively impact nematode classification via deep learning. In this paper, we propose a local structure invariance guided (LSIG) domain generalization approach to enhance the model generalization of nematode local regions in unseen domains. First, a style transfer method is introduced to synthesize new domain image samples from the source domain. Unlike in the original input images, the color information of the synthetic images is changed, but their structural information is retained. Then, a metric learning strategy is designed to determine the cross-domain invariant structural representation between the source and new domains by pairwise learning. Each class is then effectively clustered, and a better decision boundary is determined to improve the model generalization. Overall, we demonstrate the effectiveness and robustness of the method on binary-class and multi-class classification tasks on diverse nematode datasets.

A novel automated deep learning approach for Alzheimer's disease classification

Alzheimer's disease is a degenerative brain illness, incurable and progressive. Globally for every two seconds, someone is affected by Alzheimer's disease. Alzheimer's disease in the elderly is difficult to diagnose due to the complexity of the brain structure. Its pixel intensity is similar and systematic distinction is necessary. Deep learning has inspired a lot of interest in recent years in tackling challenges in a variety of fields, including medical imaging. One of the drawbacks of deep learning approach is the inability to detect changes in functional connectivity in MCI (mild cognitive impairment) patients' functional brain networks. In this paper, we utilize deep features extracted from two pre-trained deep learning models to tackle this issue. The proposed models DenseNet121 and MobileNetV2 is used to perform the task of Alzheimer's disease multi-class classification. In this method, initially we increased 70 % of dataset and generated images by using CycleGAN (generative adversarial networks). We achieved 98.82% of accuracy with proposed models. It gives best results compared to existing models.

Exploiting domain knowledge to address class imbalance and a heterogeneous feature space in multi-class classification

Real-world data of multi-class classification tasks often show complex data characteristics that lead to a reduced classification performance. Major analytical challenges are a high degree of multi-class imbalance within data and a heterogeneous feature space, which increases the number and complexity of class patterns. Existing solutions to classification or data pre-processing only address one of these two challenges in isolation. We propose a novel classification approach that explicitly addresses both challenges of multi-class imbalance and heterogeneous feature space together. As main contribution, this approach exploits domain knowledge in terms of a taxonomy to systematically prepare the training data. Based on an experimental evaluation on both real-world data and several synthetically generated data sets, we show that our approach outperforms any other classification technique in terms of accuracy. Furthermore, it entails considerable practical benefits in real-world use cases, e.g., it reduces rework required in the area of product quality control.

UHated: hate speech detection in Urdu language using transfer learning

Social media has become a driving force for social change in the global society. Events that take place in one part of the world can quickly reverberate across the globe due to the vast amount of data generated on these platforms. However, developers of these platforms face numerous challenges in keeping cyberspace as inclusive and healthy as possible. In recent years, there has been an increase in offensive and hate speech on social media. Manual efforts to address this issue have been inadequate due to the vast scope of the problem. Therefore, there is a need for an automated technique that can detect and remove offensive and hateful comments before they can cause harm. In this research, we use transfer learning to utilize pre-trained FastText Urdu word embeddings and multi-lingual BERT embeddings (RoBERTa) for our task. We also develop an Urdu language hate lexicon and use it to create an annotated dataset of 7800 Urdu tweets. Our results show that RoBERTa is able to achieve a macro F1-score of 0.82 on our multi-class classification task, outperforming deep learning and machine learning baseline models.

Deep Learning-Based Intra Mode Derivation for Versatile Video Coding

In intra coding, Rate Distortion Optimization (RDO) is performed to achieve the optimal intra mode from a pre-defined candidate list. The optimal intra mode is also required to be encoded and transmitted to the decoder side besides the residual signal, where lots of coding bits are consumed. To further improve the performance of intra coding in Versatile Video Coding (VVC) , an intelligent intra mode derivation method is proposed in this paper, termed as Deep Learning based Intra Mode Derivation (DLIMD) . In specific, the process of intra mode derivation is formulated as a multi-class classification task, which aims to skip the module of intra mode signaling for coding bits reduction. The architecture of DLIMD is developed to adapt to different quantization parameter settings and variable coding blocks including non-square ones, where only one single trained model is required. Different from the existing deep learning based classification problems, the hand-crafted features are also fed into intra mode derivation network besides the learned features from feature learning network. To compete with traditional methods, one additional binary flag is utilized in the video codec to indicate the selected scheme with RDO. Extensive experimental results reveal that the proposed method can achieve 2.28%, 1.74%, and 2.18% bit rate reduction on average for Y, U, and V components on the platform of VVC test model, which outperforms the state-of-the-art works.

ISE-Hate: A benchmark corpus for inter-faith, sectarian, and ethnic hatred detection on social media in Urdu

Social media has become the most popular platform for free speech. This freedom of speech has given opportunities to the oppressed to raise their voice against injustices, but on the other hand, this has led to a disturbing trend of spreading hateful content of various kinds. Pakistan has been dealing with the issue of sectarian and ethnic violence for the last three decades and now due to freedom of speech, there is a growing trend of disturbing content about religion, sect, and ethnicity on social media. This necessitates the need for an automated system for the detection of controversial content on social media in Urdu which is the national language of Pakistan. The biggest hurdle that has thwarted the Urdu language processing is the scarcity of language resources, annotated datasets, and pretrained language models. In this study, we have addressed the problem of detecting Interfaith, Sectarian, and Ethnic hatred on social media in Urdu language using machine learning and deep learning techniques. In particular, we have: (1) developed and presented guidelines for annotating Urdu text with appropriate labels for two levels of classification, (2) developed a large dataset of 21,759 tweets using the developed guidelines and made it publicly available, and (3) conducted experiments to compare the performance of eight supervised machine learning and deep learning techniques, for the automated identification of hateful content. In the first step, experiments are performed for the hateful content detection as a binary classification task, and in the second step, the classification of Interfaith, Sectarian and Ethnic hatred detection is performed as a multiclass classification task. Overall, Bidirectional Encoder Representation from Transformers (BERT) proved to be the most effective technique for hateful content identification in Urdu tweets.

SECL: Sampling enhanced contrastive learning

Instance-level contrastive learning such as SimCLR has been successful as a powerful method for representation learning. However, SimCLR suffers from problems of sampling bias, feature bias and model collapse. A set-level based Sampling Enhanced Contrastive Learning (SECL) method based on SimCLR is proposed in this paper. We use the proposed super-sampling method to expand the augmented samples into a contrastive-positive set, which can learn class features of the target sample to reduce the bias. The contrastive-positive set includes Augmentations (the original augmented samples) and Neighbors (the super-sampled samples). We also introduce a samples-correlation strategy to prevent model collapse, where a positive correlation loss or a negative correlation loss is computed to adjust the balance of model’s Alignment and Uniformity. SECL reaches 94.14% classification precision on SST-2 dataset and 89.25% on ARSC dataset. For the multi-class classification task, SECL achieves 90.99% on AGNews dataset. They are all about 1% higher than the precision of SimCLR. Experiments show that the training convergence of SECL is faster, and SECL reduces the risk of bias and model collapse.

Detection and Grade Classification of Diabetic Retinopathy and Adult Vitelliform Macular Dystrophy Based on Ophthalmoscopy Images

Diabetic retinopathy (DR) and adult vitelliform macular dystrophy (AVMD) may cause significant vision impairment or blindness. Prompt diagnosis is essential for patient health. Photographic ophthalmoscopy checks retinal health quickly, painlessly, and easily. It is a frequent eye test. Ophthalmoscopy images of these two illnesses are challenging to analyse since early indications are typically absent. We propose a deep learning strategy called ActiveLearn to address these concerns. This approach relies heavily on the ActiveLearn Transformer as its central structure. Furthermore, transfer learning strategies that are able to strengthen the low-level features of the model and data augmentation strategies to balance the data are incorporated owing to the peculiarities of medical pictures, such as their limited quantity and generally rigid structure. On the benchmark dataset, the suggested technique is shown to perform better than state-of-the-art methods in both binary and multiclass accuracy classification tasks with scores of 97.9% and 97.1%, respectively.

Event Detection via Context Understanding Based on Multi-task Learning

Event detection (ED) aims to identify events of interest described in the text. With the current explosive growth of text data on the internet, ED is increasingly practical and has gained many researchers’ attention. The existing works usually design ED as a token-level multi-class classification task. In this setting, given a sentence, ED models’ prediction for each token is relatively independent and thus cannot fully utilize sentence-level information and the association relations between multiple events in this sentence. To handle these situations, this paper proposes a multi-task learning based event detection model, which introduces an event type oriented text classification as an auxiliary task to improve the model’s understanding of sentence-level information. In addition, this model utilizes a Conditional Random Field (CRF) to explore the correlations between various event types and constrain the model’s output space. Experimental comparisons with state-of-the-art baselines on DuEE dataset demonstrate the model’s effectiveness.