Perform Data Augmentation Research Articles

D2BOF-COVIDNet: A Framework of Deep Bayesian Optimization and Fusion-Assisted Optimal Deep Features for COVID-19 Classification Using Chest X-ray and MRI Scans.

In 2019, a corona virus disease (COVID-19) was detected in China that affected millions of people around the world. On 11 March 2020, the WHO declared this disease a pandemic. Currently, more than 200 countries in the world have been affected by this disease. The manual diagnosis of this disease using chest X-ray (CXR) images and magnetic resonance imaging (MRI) is time consuming and always requires an expert person; therefore, researchers introduced several computerized techniques using computer vision methods. The recent computerized techniques face some challenges, such as low contrast CTX images, the manual initialization of hyperparameters, and redundant features that mislead the classification accuracy. In this paper, we proposed a novel framework for COVID-19 classification using deep Bayesian optimization and improved canonical correlation analysis (ICCA). In this proposed framework, we initially performed data augmentation for better training of the selected deep models. After that, two pre-trained deep models were employed (ResNet50 and InceptionV3) and trained using transfer learning. The hyperparameters of both models were initialized through Bayesian optimization. Both trained models were utilized for feature extractions and fused using an ICCA-based approach. The fused features were further optimized using an improved tree growth optimization algorithm that finally was classified using a neural network classifier. The experimental process was conducted on five publically available datasets and achieved an accuracy of 99.6, 98.5, 99.9, 99.5, and 100%. The comparison with recent methods and t-test-based analysis showed the significance of this proposed framework.

A pyramid input augmented multi-scale CNN for GGO detection in 3D lung CT images

This paper proposes a new convolutional neural network (CNN) with multi-scale processing for detecting ground-glass opacity nodules (GGO) in 3D computed tomography (CT) images, which is referred to as PiaNet for short. PiaNet consists of a feature-extraction module and a prediction module. The former module is constructed by introducing pyramid multi-scale source connections into a contracting-expanding structure. Besides, a new multi-receptive-field convolution block (MRCB) is presented to fuse the convolutions with multiple kernels of varying sizes for capturing features in each scale of information better. The latter module includes a bounding-box regressor and a classifier that are employed to simultaneously recognize GGO nodules and estimate bounding boxes at multiple scales. To train the proposed PiaNet, a two-stage transfer learning strategy is developed. In the first stage, the feature-extraction module is embedded into a classifier network that is trained on a large data set of GGO and non-GGO patches, which are generated by performing data augmentation from a small number of annotated CT scans. In the second stage, the pretrained feature-extraction module is loaded into PiaNet, and then PiaNet is fine-tuned using the annotated CT scans. We evaluate the proposed PiaNet with the LIDC-IDRI dataset. The experimental results demonstrate that our method outperforms state-of-the-art counterparts, including the Subsolid CAD and Aidence systems and CPM-Net and S4ND and GA-SSD methods. PiaNet achieves a sensitivity of 93.6% with only one false positive per scan.

Applying an Intelligent Approach to Environmental Sustainability Innovation in Complex Scenes

Environmental protection is still a key issue that cannot be ignored at this stage of social development. With the development of artificial intelligence, various technologies increasingly tend to be widely used in the field of environmental protection, such as searching the wilderness through an unmanned aerial vehicle (UAV) and cleaning garbage by robots. Traditional object detection algorithms for this scenario suffer from low accuracy and high computational cost. Therefore, this paper proposes an algorithm applied to automatic garbage detection and instance segmentation in complex scenes. First, we construct sample-fused feature pyramid networks (SF-FPN) to achieve multi-scale feature sampling on multiple levels, to enhance the semantic representation of features. Second, adding the mask branch based on conditional convolution, introducing the idea of instance-filters to automatically generate the filter parameters of the Fully Convolutional Networks (FCN), to realize the instance-level pixel classification. Moreover, the Atrous Spatial Pyramid Pooling (ASPP) module is introduced to encode the feature information in a dense way to assist the generation of MASK. Finally, the object is detected and the instance is segmented by a two-branch structure. In addition, we also perform data augmentation on the original dataset to prevent model overfitting. The proposed algorithm reaches 82.7 and 72.4 according to the mAP index of detection and instance segmentation while using the public TACO dataset.

An End-to-End Contrastive Self-Supervised Learning Framework for Language Understanding

Abstract Self-supervised learning (SSL) methods such as Word2vec, BERT, and GPT have shown great effectiveness in language understanding. Contrastive learning, as a recent SSL approach, has attracted increasing attention in NLP. Contrastive learning learns data representations by predicting whether two augmented data instances are generated from the same original data example. Previous contrastive learning methods perform data augmentation and contrastive learning separately. As a result, the augmented data may not be optimal for contrastive learning. To address this problem, we propose a four-level optimization framework that performs data augmentation and contrastive learning end-to-end, to enable the augmented data to be tailored to the contrastive learning task. This framework consists of four learning stages, including training machine translation models for sentence augmentation, pretraining a text encoder using contrastive learning, finetuning a text classification model, and updating weights of translation data by minimizing the validation loss of the classification model, which are performed in a unified way. Experiments on datasets in the GLUE benchmark (Wang et al., 2018a) and on datasets used in Gururangan et al. (2020) demonstrate the effectiveness of our method.

A machine learning-based framework for forecasting sales of new products with short life cycles using deep neural networks

Demand forecasting is becoming increasingly important as firms launch new products with short life cycles more frequently. This paper provides a framework based on state-of-the-art techniques that enables firms to use quantitative methods to forecast sales of newly launched, short-lived products that are similar to previous products when there is limited availability of historical sales data for the new product. In addition to exploiting historical data using time-series clustering, we perform data augmentation to generate sufficient sales data and consider two quantitative cluster assignment methods. We apply one traditional statistical (ARIMAX) and three machine learning methods based on deep neural networks (DNNs) – long short-term memory, gated recurrent units, and convolutional neural networks. Using two large data sets, we investigate the forecasting methods’ comparative performance and, for the larger data set, show that clustering generally results in substantially lower forecast errors. Our key empirical finding is that simple ARIMAX considerably outperforms the more advanced DNNs, with mean absolute errors up to 21%–24% lower. However, when adding Gaussian white noise in our robustness analysis, we find that ARIMAX’s performance deteriorates dramatically, whereas the considered DNNs display robust performance. Our results provide insights for practitioners on when to use advanced deep learning methods and when to use traditional methods.

Synthetic Data Generation By Artificial Intelligence to Accelerate Translational Research and Precision Medicine in Hematological Malignancies

Background In hematological malignancies there is a growing demand for real-world, comprehensive data including clinical and genomic information to build powerful models to improve diagnosis, prognosis and personalized treatment choice. However, collecting such information in large patient populations is challenging and there are many issues concerning patient privacy that need to be accounted for. One approach that can circumvent these issues is the creation of synthetic data that captures the complexities of the original data set (distributions, non-linear relationships, and noise) without including any real patient information. Aims 1) Apply advanced synthetic data generation methods to real-world datasets of different hematological malignancies. 2) Develop a Synthetic Validation Framework to evaluate the quality of synthetic data and perform data augmentation. 3) Test the capability of synthetic data to accelerate translational research. Methods Here we implemented a Conditional Tabular Wasserstein Generative Adversarial Networks (GAN) architecture with Gradient Penalty to generate synthetic data. Use cases were different cohorts of patients with myelodysplastic syndrome (MDS) and acute myeloid leukemia (AML) with available clinical and molecular features. We created a Synthetic Validation Framework to evaluate the quality of generated synthetic data: Clinical Synthetic Fitness (CSF) and Genomic Synthetic Fitness (GSF) scores were calculated as the average of multiple metric tests adopted. Patients were stratified by Hierarchical Dirichlet (HD) clustering. Explainability analysis was carried out by SHapley Additive exPlanations approach (SHAP). Survival analyses were performed by Kaplan-Meier curves and CoxPH models (Experimental plan is reported in Figure 1). Results We first created a synthetic copy of a MDS cohort (n=2,043) using all the real data for training the model. We compared synthetic vs. real data, obtaining high fitness performances for both clinical and genomic features (CSF=93%; GSF=90%). HD were then applied to define clusters capturing broad dependencies among genomic features, showing comparable results in synthetic vs. real data; SHAP analysis indicated that similar features drive patients’ classification in both datasets. Finally, synthetic patients had comparable survival with respect to real ones; when applying conventional scoring system (IPSS-R), the probability of survival of the 5 risk categories was comparable between synthetic and real data. In the second experiment setting, we analysed synthetic MDS datasets with different size generated with model trained on a real dataset. Interestingly, when generating a synthetic augmented dataset (200%) we obtained high fitness performance for both clinical and genomic features (CSF=91%; GSF=89%). Moreover, all the performances showed a similar trend when considering a cohort of 1,002 patients with AML (CSF=92%; GSF=89%) thus proving evidence for high generalizability of the model across different clinical settings. Finally, we investigated if the generation of synthetic data can accelerate translational research in hematology. Since the first publication on clinical relevance of gene mutations in MDS (Leukemia 2014;28:241), it took several years to collect data in large patient populations for generating a molecular classification (JCO 2021;39:1223) and prognostic score (IPSS-M, NEJM Evid 2022;1:7). Starting from the MDS cohort available in 2014 (n=944, Leukemia 2014;28:241), we generated 300% augmented synthetic dataset. HD were applied to synthetic data to define genomic-based clinical entities, resulting in the identification of the same 8 subgroups described in a real cohort of 2,043 patients many years later. Moreover, we applied a CoxPH model to the synthetic dataset to generate a molecular prognostic score (IPSS-M_Syn). The model was based on similar molecular features as the real IPSS-M and identified 6 risk categories in which the probability of survival was similar to that of IPSS-M risk groups (Figure 2). Conclusion GAN-generated synthetic data recapitulate statistical properties and complexity of clinical and genomic features in different hematological malignancies, replicate reliable survival estimates and allow effective data augmentation. The implementation of this technology seems to accelerate precision medicine research in hematology. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

An explainable COVID-19 detection system based on human sounds.

Acoustic signals generated by the human body have often been used as biomarkers to diagnose and monitor diseases. As the pathogenesis of COVID-19 indicates impairments in the respiratory system, digital acoustic biomarkers of COVID-19 are under investigation. In this paper, we explore an accurate and explainable COVID-19 diagnosis approach based on human speech, cough, and breath data using the power of machine learning. We first analyze our design space considerations from the data aspect and model aspect. Then, we perform data augmentation, Mel-spectrogram transformation, and develop a deep residual architecture-based model for prediction. Experimental results show that our system outperforms the baseline, with the ROC-AUC result increased by 5.47%. Finally, we perform an interpretation analysis based on the visualization of the activation map to further validate the model.

Research on the Guidance of Youth Labor Education Based on the "Combination of Education and Production Labor" Program Based on the Deep Learning Model.

At present, there is a lack of research on Marx's idea of “combining education and productive labor” and its guiding significance for youth labor education, and no effective teaching model has been formed. In response to this problem, this study proposes a semi-supervised deep learning model based on u-wordMixup (SD-uwM). When there is a shortage of labeled samples, semi-supervised learning uses a large number of unlabeled samples to solve the problem of labeling bottlenecks. However, since the unlabeled samples and labeled samples come from different fields, there may be quality problems in the unlabeled samples, which makes the generalization ability of the model worse., resulting in a decrease in classification accuracy. The model uses the u-wordMixup method to perform data augmentation on unlabeled samples. Under the constraints of supervised cross-entropy and unsupervised consistency loss, it can improve the quality of unlabeled samples and reduce overfitting. The comparative experimental results on the AGNews, THUCNews, and 20Newsgroups data sets show that the proposed method can improve the generalization ability of the model and also effectively improve the time performance. The study found that the SD-uwM model uses the u-wordMixup method to enhance the unlabeled samples and combines the idea of the Mean Teacher model, which can significantly improve the text classification performance. The SD-uwM model can improve the generalization ability and time performance of the model, respectively, 86.4 ± 1.3 and 90.5 ± 1.3. Therefore, the use of SD-uwM in Marx's program is of great practical significance for the guidance process of youth labor education.

A disaster classification application using convolutional neural network by performing data augmentation

Natural disasters are catastrophic events and cause havoc to human life. These events occur in the most unpredictable times and are beyond human control. The aftermath of the disasters is devastating ranging from loss of life to relocation of large groups of the population. With the development in the domains of computer vision (CV) and Image processing, machine learning and deep learning models can integrate images and perform predictions. Deep learning techniques employ many robust techniques and provide significant results even in the case of images. The detection of natural disasters without human intervention requires the help of deep learning techniques. The project aims to employ a multi-layered convolutional neural network (CNN) organization to classify the images related to natural disasters related to earthquakes, floods, cyclones, and wildfires.

On the analysis of data augmentation methods for spectral imaged based heart sound classification using convolutional neural networks

BackgroundThe application of machine learning to cardiac auscultation has the potential to improve the accuracy and efficiency of both routine and point-of-care screenings. The use of convolutional neural networks (CNN) on heart sound spectrograms in particular has defined state-of-the-art performance. However, the relative paucity of patient data remains a significant barrier to creating models that can adapt to a wide range of potential variability. To that end, we examined a CNN model’s performance on automated heart sound classification, before and after various forms of data augmentation, and aimed to identify the most optimal augmentation methods for cardiac spectrogram analysis.ResultsWe built a standard CNN model to classify cardiac sound recordings as either normal or abnormal. The baseline control model achieved a PR AUC of 0.763 ± 0.047. Among the single data augmentation techniques explored, horizontal flipping of the spectrogram image improved the model performance the most, with a PR AUC of 0.819 ± 0.044. Principal component analysis color augmentation (PCA) and perturbations of saturation-value (SV) of the hue-saturation-value (HSV) color scale achieved a PR AUC of 0.779 ± 045 and 0.784 ± 0.037, respectively. Time and frequency masking resulted in a PR AUC of 0.772 ± 0.050. Pitch shifting, time stretching and compressing, noise injection, vertical flipping, and applying random color filters negatively impacted model performance. Concatenating the best performing data augmentation technique (horizontal flip) with PCA and SV perturbations improved model performance.ConclusionData augmentation can improve classification accuracy by expanding and diversifying the dataset, which protects against overfitting to random variance. However, data augmentation is necessarily domain specific. For example, methods like noise injection have found success in other areas of automated sound classification, but in the context of cardiac sound analysis, noise injection can mimic the presence of murmurs and worsen model performance. Thus, care should be taken to ensure clinically appropriate forms of data augmentation to avoid negatively impacting model performance.

RG-GCN: A Random Graph Based on Graph Convolution Network for Point Cloud Semantic Segmentation

Point cloud semantic segmentation, a challenging task in 3D data processing, is popular in many realistic applications. Currently, deep learning methods are gradually being applied to point cloud semantic segmentation. However, as it is difficult to manually label point clouds in 3D scenes, it remains difficult to obtain sufficient training samples for the supervised deep learning network. Although an increasing number of excellent methods have been proposed in recent years, few of these have focused on the problem of semantic segmentation with insufficient samples. To address this problem, this paper proposes a random graph based on graph convolution network, referred to as RG-GCN. The proposed network consists of two key components: (1) a random graph module is proposed to perform data augmentation by changing the topology of the built graphs; and (2) a feature extraction module is proposed to obtain local significant features by aggregating point spatial information and multidimensional features. To validate the performance of the RG-GCN, the indoor dataset S3DIS and outdoor dataset Toronto3D are used to validate the proposed network via a series of experiments. The results show that the proposed network achieves excellent performance for point cloud semantic segmentation of the two different datasets.

Data augmentation strategies for EEG-based motor imagery decoding

The wide use of motor imagery as a paradigm for brain-computer interfacing (BCI) points to its characteristic ability to generate discriminatory signals for communication and control. In recent times, deep learning techniques have increasingly been explored, in motor imagery decoding. While deep learning techniques are promising, a major challenge limiting their wide adoption is the amount of data available for decoding. To combat this challenge, data augmentation can be performed, to enhance decoding performance. In this study, we performed data augmentation by synthesizing motor imagery (MI) electroencephalography (EEG) trials, following six approaches. Data generated using these methods were evaluated based on four criteria, namely – the accuracy of prediction, the Frechet Inception distance (FID), the t-distributed Stochastic Neighbour Embedding (t-SNE) plots and topographic head plots. We show, based on these, that the synthesized data exhibit similar characteristics with real data, gaining up to 3% and 12% increases in mean accuracies across two public datasets. Finally, we believe these approaches should be utilized in applying deep learning techniques, as they not only have the potential to improve prediction performances, but also to save time spent on subject data collection.

Deep Convolutional Neural Networks For Environmental Sound Classification

Abstract: We propose a model to classify environmental sounds such as People Sounds, Vehicles Sounds, Siren Sounds, Horn, Engine Sounds. We perform Data Augmentation techniques to extract best features from the given audio to classify which class of sound. Our deep convolutional neural network architecture uses stacked convolutional and pooling layers to extract highlevel feature representations from spectrogram-like features from the given input.

A Deep Convolutional Generative Adversarial Networks-Based Method for Defect Detection in Small Sample Industrial Parts Images

Online defect detection in small industrial parts is of paramount importance for building closed loop intelligent manufacturing systems. However, high-efficiency and high-precision detection of surface defects in these manufacturing systems is a difficult task and poses a major research challenge. The small sample size of industrial parts available for training machine learning algorithms and the low accuracy of computer vision-based inspection algorithms are the bottlenecks that restrict the development of efficient online defect detection technology. To address these issues, we propose a small sample gear face defect detection method based on a Deep Convolutional Generative Adversarial Network (DCGAN) and a lightweight Convolutional Neural Network (CNN) in this paper. Initially, we perform data augmentation by using DCGAN and traditional data enhancement methods which effectively increase the size of the training data. In the next stage, we perform defect classification by using a lightweight CNN model which is based on the state-of-the-art Vgg11 network. We introduce the Leaky ReLU activation function and a dropout layer in the proposed CNN. In the experimental evaluation, the proposed framework achieves a high score of 98.40%, which is better than that of the classic Vgg11 network model. The method proposed in this paper is helpful for the detection of defects in industrial parts when the available sample size for training is small.

Does Minority Case Sampling Improve Performance with Imbalanced Outcomes in Psychological Research?

In psychological research, class imbalance in binary outcome variables is a common occurrence, particularly in clinical variables (e.g., suicide outcomes). Class imbalance can present a number of difficulties for inference and prediction, prompting the development of a number of strategies that perform data augmentation through random sampling from just the positive cases, or from both the positive and negative cases. Through evaluation in benchmark datasets from computer science, these methods have shown marked improvements in predictive performance when the outcome is imbalanced. However, questions remain regarding generalizability to psychological data. To study this, we implemented a simulation study that tests a number of popular sampling strategies implemented in easy-to-use software, as well as in an empirical example focusing on the prediction of suicidal thoughts. In general, we found that while one sampling strategy demonstrated far worse performance even in comparison to no sampling, the other sampling methods performed similarly, evidencing slight improvements over no sampling. Further, we evaluated the sampling strategies across different forms of cross-validation, model fit metrics, and machine learning algorithms.

Artificial intelligence based detection of age-related macular degeneration using optical coherence tomography with unique image preprocessing.

The aim of the study is to improve the accuracy of age related macular degeneration (AMD) disease in its earlier phases with proposed Capsule Network (CapsNet) architecture trained on speckle noise reduced spectral domain optical coherence tomography (SD-OCT) images based on an optimized Bayesian non-local mean (OBNLM) filter augmentation techniques. A total of 726 local SD-OCT images were collected and labelled as 159 drusen, 145 dry AMD, 156 wet AMD and 266 normal. Region of interest (ROI) was identified. Speckle noise in SD-OCT images were reduced based on OBNLM filter. The processed images were fed to proposed CapsNet architecture to clasify SD-OCT images. Accuracy rates were calculated in both public and local dataset. Accuracy rate of local SD-OCT image dataset classification was achieved to a value of 96.39% after performing data augmentation and speckle noise reduction with OBNLM. The performance of proposed CapsNet was also evaluated on the public Kaggle dataset under the same processing procedures and the accuracy rate was calculated as 98.07%. The sensitivity and specificity rates were 96.72% and 99.98%, respectively. The classification success of proposed CapsNet may be improved with robust pre-processing steps like; determination of ROI and denoised SD-OCT images based on OBNLM. These impactful image preprocessing steps yielded higher accuracy rates for determining different types of AMD including its precursor lesion on the both local and public dataset with proposed CapsNet architecture.

A Multi-Level Optimization Framework for End-to-End Text Augmentation

AbstractText augmentation is an effective technique in alleviating overfitting in NLP tasks. In existing methods, text augmentation and downstream tasks are mostly performed separately. As a result, the augmented texts may not be optimal to train the downstream model. To address this problem, we propose a three-level optimization framework to perform text augmentation and the downstream task end-to- end. The augmentation model is trained in a way tailored to the downstream task. Our framework consists of three learning stages. A text summarization model is trained to perform data augmentation at the first stage. Each summarization example is associated with a weight to account for its domain difference with the text classification data. At the second stage, we use the model trained at the first stage to perform text augmentation and train a text classification model on the augmented texts. At the third stage, we evaluate the text classification model trained at the second stage and update weights of summarization examples by minimizing the validation loss. These three stages are performed end-to-end. We evaluate our method on several text classification datasets where the results demonstrate the effectiveness of our method. Code is available at https://github.com/Sai-Ashish/End-to-End-Text-Augmentation.

A review: Data pre-processing and data augmentation techniques

This review paper provides an overview of data pre-processing in Machine learning, focusing on all types of problems while building the machine learning problems. It deals with two significant issues in the pre-processing process (i). issues with data and (ii). Steps to follow to do data analysis with its best approach. As raw data are vulnerable to noise, corruption, missing, and inconsistent data, it is necessary to perform pre-processing steps, which is done using classification, clustering, and association and many other pre-processing techniques available. Poor data can primarily affect the accuracy and lead to false prediction, so it is necessary to improve the dataset's quality. So, data pre-processing is the best way to deal with such problems. It makes the knowledge extraction from the data set much easier with cleaning, Integration, transformation, and reduction methods. The issue with Data missing and significant differences in the variety of data always exists as the information is collected through multiple sources and from a real-world application. So, the data augmentation approach generates data for machine learning models. To decrease the dependency on training data and to improve the performance of the machine learning model. This paper discusses flipping, rotating with slight degrees and others to augment the image data and shows how to perform data augmentation methods without distorting the original data.

Pengenalan Emosi Pembicara Menggunakan Convolutional Neural Networks

Recognition of the speaker's emotions is an important but challenging component of Human-Computer Interaction (HCI). The need for the recognition of the speaker's emotions is also increasing related to the need for digitizing the company's operational processes related to the implementation of industry 4.0. The use of Deep Learning methods is currently increasing, especially for processing unstructured data such as data from voice signals. This study tries to apply the Deep Learning method to classify the speaker's emotions using an open dataset from SAVEE which contains seven classes of voice emotions in English. The dataset will be trained using the CNN model. The final accuracy of the model is 88% on the training data and 52% on the test data, which means the model is overfitting. This is due to the imbalance of emotion classes in the dataset, which makes the model tend to predict classes with more labels. In addition, the lack of heterogeneity of the dataset makes the character of the emotion class more different from the others so that it can reduce the bias in the model so as not to overfit the model. Further development of this research can be done, such as over-sampling the existing dataset by adding other data sources, then performing data augmentation to get the data character of each emotion class and setting hyperparameter values ​​to get better accuracy values.

Transfer-Learning-Based Approach for the Diagnosis of Lung Diseases from Chest X-ray Images.

Using chest X-ray images is one of the least expensive and easiest ways to diagnose patients who suffer from lung diseases such as pneumonia and bronchitis. Inspired by existing work, a deep learning model is proposed to classify chest X-ray images into 14 lung-related pathological conditions. However, small datasets are not sufficient to train the deep learning model. Two methods were used to tackle this: (1) transfer learning based on two pretrained neural networks, DenseNet and ResNet, was employed; (2) data were preprocessed, including checking data leakage, handling class imbalance, and performing data augmentation, before feeding the neural network. The proposed model was evaluated according to the classification accuracy and receiver operating characteristic (ROC) curves, as well as visualized by class activation maps. DenseNet121 and ResNet50 were used in the simulations, and the results showed that the model trained by DenseNet121 had better accuracy than that trained by ResNet50.