Search

The widespread use of social media like Facebook, Twitter, Whatsapp, etc. has changed the way News is created and published; accessing news has become easy and inexpensive. However, the scale of usage and inability to moderate the content has made social media, a breeding ground for the circulation of fake news. Fake news is deliberately created either to increase the readership or disrupt the order in the society for political and commercial benefits. It is of paramount importance to identify and filter out fake news especially in democratic societies. Most existing methods for detecting fake news involve traditional supervised machine learning which has been quite ineffective. In this paper, we are analyzing word embedding features that can tell apart fake news from true news. We use the LIAR and ISOT data set. We churn out highly correlated news data from the entire data set by using cosine similarity and other such metrices, in order to distinguish their domains based on central topics. We then employ auto-encoders to detect and differentiate between true and fake news while also exploring their separability through network analysis.

Deep Learning plays a major role in advancements in the healthcare domain, of which early disease diagnosis is one main application. With respect to the same, deep learning-based classification of brain MRI at the subject level is the requirement in the medical field. In this work, we have implemented an algorithm to identify the most discriminative range of MRI slices at the subject level to differentiate between Normal Cohorts (NC) and Parkinson’s Disease (PD) subjects. We have also focused on handling data leakage and verified the model generalizability using Stratified k-fold cross-validation.

Diabetic retinopathy (DR) is a complication of diabetes which is due to the impairment of blood vessels of photosensitive cells in the eyes. This complication results in loss of eyesight if not diagnosed in the early stages. There are five stages of diabetic retinopathy: No DR, mild, moderate, severe, and proliferative. DR detection by traditional methods consumes a lot of time. An automatic and precise model would require an adequate amount of data for training which is not available. Publicly available dataset is highly imbalanced for other classes apart from No DR, especially proliferative and severe classes. In this paper, a model is created in two stages. The first stage involves the generation of synthetic data points using Deep Convolutional Generative Adversarial Network (DCGAN). The synthetic data are for highly imbalanced classes namely severe and proliferative. The second phase involves the augmented data classification using a CNN architecture.

Diagnosis is the key step forward to cure a disease. Deep learning is becoming popular as a tool for usage in medical diagnosis. The existing literature using deep learning for the diagnosis of Parkinson’s disease (PD) by transfer learning of MRI data was limited to the AlexNet architecture. The present work aims to inculcate commonly used deep learning architectures using transfer learning for effective diagnosis of PD using MRI data. The best three performing models are selected based on the standard metric called F1-score. An ensemble model is proposed based on the maximum probability across all the selected models for PD classification. The approach mainly focuses on the effective diagnosis of PD. The performance of the proposed ensemble approach is validated using the standard metrics known as F1-score and classification accuracy. Among the commonly used deep learning architectures, it was found that VGG19 is better than the existing state-of-the-art, which is AlexNet. The proposed ensemble approach applied on top three commonly used deep learning models led to the improvement in accuracy to 0.978. The false positive (FP) and false negative (FN) were reduced significantly by the proposed ensemble approach.

Accurate automatic Identification and localization of spine vertebrae points in CT scan images is crucial in medical diagnosis. This paper presents an automatic feature extraction network, based on transfer learned CNN, in order to handle the availability of limited samples. The 3D vertebrae centroids are identified and localized by an LSTM network, which is trained on CNN features extracted from 242 CT spine sequences. The model is further trained to estimate age and gender from LSTM features. Thus, we present a framework that serves as a multi-task data driven model for identifying and localizing spine vertebrae points, age estimation and gender classification. The proposed approach is compared with benchmark results obtained by testing 60 scans. The advantage of the multi-task framework is that it does not need any additional information other than the annotations on the spine images indicating the presence of vertebrae points.

The price of a single stock is seldom independent. It has been known to brokers and fund managers, that, they heavily influence each other. Portfolios are built, on the premise of minimizing such dependencies between stocks. There have been several efforts to quantify these dependencies, predominantly using conventional statistics and correlations. Most approaches use single independent variable-based approaches to compare two stock price signals at a time. In this paper, we analyze stock sectors with multiple stocks in each, presenting a multivariate case. We apply two multivariate analysis techniques, namely multivariate variational mode decomposition (MVMD) and spatio-temporal intrinsic mode decomposition (STIMD), to analyze stock sectors based on their individual stock day-wise price series. Sector-wise data is downloaded from Google Finance. Furthermore, we quantify the dependence of a sector on each of its constituent stock, by decomposing this multivariate signal and selectively reconstructing it multiple times, excluding each one from the reconstruction process. The error of reconstruction in each case serves as a measure of how much other stocks depend on the one that was excluded.

One of the most common causes of failures in complex systems in nature or engineering is an abrupt transition from a stable to an alternate stable state. Such transitions cause failures in the dynamic power systems. We focus on this transition from a stable to an unstable manifold for a rate-dependent mechanical power input via a numerical investigation in a theoretical power system model. Our studies uncover early transitions that depend on the rate of variation of mechanical input. Furthermore, we determine the dependency of a critical rate on initial conditions of the system. Accordingly, this knowledge of the critical rate can be used in devising an effective control strategy based on artificial intelligence (AI).

Multiple myeloma cancer is caused by the abnormal growth of plasma cells in the bone marrow. The most commonly used method for diagnosis of multiple myeloma is Bone marrow aspiration, where the aspirate slide images are either observed visually or passed onto existing digital image processing software for the detection of myeloma cells. The current work explores the effectiveness of deep learning based object detection/segmentation algorithms such as Mask-RCNN and unet for the detection of multiple myeloma. The manual polygon annotation of the current dataset is performed using VGG image annotation software. The deep learning models were trained by monitoring the train and validation loss per epoch and the best model was selected based on the minimal loss for the validation data. From the comparison results obtained for both the models, it is observed that Mask-RCNN has competing results than unet and it addresses most of the challenges existing in multiple myeloma segmentation.

Diabetic Retinopathy(DR) is one of the diseases, which is caused by damage to blood vessels of the light sensitive tissue of the eyes. DR is graded into five different levels; normal, mild, moderate, severe and proliferative. DR diagnosis demands more time on detection from fundus images. An accurate automatic model requires sufficient data for training which is unavailable. The open-source DR datasets are highly imbalanced between the levels of DR and it is also not easy to collect more data for proliferative cases. The synthetic generation of data for such highly imbalanced classes in a dataset provides better results on classification. In this paper, an analysis of classification results for the same is carried out with augmented proliferative class (highly imbalanced class) in the EYEPACS dataset using a generative adversarial network (GAN). We generated highly diverse images for proliferative cases without any constraints. The generated proliferative images do not influence other class images and have also improved the classification results obtained by the model, over that which was trained without synthetic generation. The results obtained before and after augmentation by the proposed generative based model is compared over various model attributes.

The neural network, one of the intelligent data mining technique that has been used by researchers in various areas for the past 10 years. Prediction and analysis of stock market data have got an important role in today’s economy. The various algorithms used for forecasting can be categorized into linear (AR, MA, ARIMA, ARMA) and non-linear models (ARCH, GARCH, Neural Network). In this paper, we are using four types of deep learning architectures i.e Multilayer Perceptron (MLP), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) for predicting the stock price of a company based on the historical prices available. Here we are using day-wise closing price of two different stock markets, National Stock Exchange (NSE) of India and New York Stock Exchange (NYSE). The network was trained with the stock price of a single company from NSE and predicted for five different companies from both NSE and NYSE. It has been observed that CNN is outperforming the other models. The network was able to predict for NYSE even though it was trained with NSE data. This was possible because both the stock markets share some common inner dynamics. The results obtained were com- pared with ARIMA model and it has been observed that the neural networks are outperforming the existing linear model (ARIMA).

Cardiovascular disease is a condition that causes damage to the heart muscle, valves, rhythm, or blockage in the blood vessels. It requires early diagnosis, as it is the leading cause for the sudden death in humans. Electrocardiogram (ECG) is the most important biomedical signal used extensively by the cardiologist to diagnose cardiovascular disease. The classification of ECG signals helps physicians to make decisions in the diagnosis of cardiac diseases. There are many conventional machine learning and deep learning algorithms used in the literature for the automatic classification of ECG signals. Conventional machine learning algorithms require handcrafted features. There are many features such as morphological feature extraction, computation of RR interval, QRS peak detection, ST segment, ST distance, and amplitude computation. The classical machine learning algorithms used to classify the extracted features are shallow neural network, K nearest neighbor, support vector machines (SVM), random forest, and decision tree. Deep learning algorithms learn the features from the given training data, which outperformed the handcrafted features used in the conventional machine learning algorithms. There are state-of-the-art architectures such as convolutional neural network (CNN), recurrent neural network (RNN), long short-term memory (LSTM), and gated recurrent unit (GRU). The existing architectures rely on disease-specific approach. This chapter aims to provide a single architecture by transferable approach for cardiac disease classification using ECG. We plan to achieve this by the analysis of the different state-of-the-art deep learning architectures such as CNN, LSTM, RNN, and GRU to classify cardiac diseases using ECG signal. Arrhythmia, myocardial infarction, and atrial fibrillation are the cardiac diseases we considered for the study. We have used publically available datasets from Kaggle and PhysioNet for experimental evaluation. In the proposed methodology, the deep learning architectures such as RNN, LSTM, and GRU trained to classify the atrial fibrillation from the ECG signal were used to classify other cardiac diseases such as arrhythmia and myocardial infarction. Similarly, CNN, which is trained to classify arrhythmia and myocardial infarction, is used to classify atrial fibrillation. We kept the network parameters (also known as hyperparameters) such as learning rate, batch size, and number of epochs the same for the entire experimental analysis. We evaluated the performance of the proposed methodology using the metrics: precision, recall, and F1 score. We observed that LSTM and GRU performed well compared to the RNN and CNN. LSTM and GRU offer the maximum precision and recall score, which varies between 0.97–0.98 for all three diseases. The computational cost of GRU is less compared to the RNN. Our results show that the deep learning architectures are transferable. Unlike deep learning, which is the data-driven approach, the machine learning algorithms are not adaptable or transferable. To validate this, comparison between the different machine learning algorithms like Naive Bayes, K-nearest neighbor, SVMs with two different kernels such as linear and RBF, AdaBoost, random forest, decision tree, and logistic regression are conducted for all three diseases considered for the study.

Defects in structures will affect its natural vibrations. With the advent of pure data-driven modeling techniques such as Dynamic Mode Decomposition (DMD), the defected modes can be separated from the normal modes by using vibration data from various points on the structural element. In this work we simulate the vibrations of a cantilever beam in Abaqus® without defect and with different defects. We apply DMD to compute the spatial modes of vibration in each of these cases. Furthermore we train a Support Vector Machine (SVM) classifier with the Eigen-modes we have computed, to identify defects. We also analyze this data visually using t-SNE plots.

Medical diagnosis is the process of determining a patient’s health condition by the observation of symptoms and test results. Cardiovascular diseases are one of the most common causes of death worldwide. Electrocardiogram (ECG) is one of the effective ways for diagnosing heart conditions. ECG records and detects the strength and timing of electrical activity of the heart. A proper diagnosis can reduce mortality rate. Artificial intelligence (AI) has shown its inexplicable contribution in the field of medical science, especially in diagnosis. Convolutional neural network (CNN) is the most popular deep learning algorithm, which captures the relevant features by itself. Deep learning requires a massive amount of data to train the network, which increases the computational complexity. This chapter aims to reduce the computational complexity. We consider the cardiac diseases such as arrhythmia and myocardial infarction (MI) for our experimental analysis. We have used heartbeat segmented and preprocessed ECG data available at Kaggle. We aim to reduce the computational complexity of the existing deep learning architecture for cardiac disease classification by using the feature-extracted data. We propose the single-layer CNN for the classification of ECG beats of arrhythmia and MI. We also evaluated the performance of the proposed model by using the following evaluation metrics: precision, recall, and F1 score. The performance of the proposed architecture is high compared to state-of-the-art methods.

Phonocardiogram known as PCG plays a significant role in the early diagnosis of cardiac abnormalities. Phonocardiogram can be used as initial diagnostics tool in remote applications due to its simplicity and cost effectiveness. Instead of disease specific approach, the proposed work aims for the single architecture that could diagnose different cardiac abnormality from the PCG signals collected from various sources. Our study also shows the effectiveness of using Fast Fourier Transform (FFT) in signal processing applications. It avoids the trivial preprocessing and feature extraction mechanisms with the promising results.

In recent days, Artificial Intelligence is one of the emerging fields. Conventional machine learning algorithms require feature extraction, which is hand-crafted based on the required application, creates the need for the development of deep learning algorithms. In general, deep learning algorithms learn the features based on the given input data. Recently, deep learning is applied in various domains such as computer vision, speech processing, cybersecurity, healthcare, etc. The development of deep learning applications in health care domain is increasing rapidly. One of the emerging healthcare applications is cardiovascular disease classification from ECG signals using deep learning. Myocardial infarction is a predominant type of cardiovascular disease, which affects one-third of the population annually. Conventional machine learning and deep learning techniques are used for diagnosis of myocardial infarction. In this paper, we explore the deep residual CNN, which is designed for arrhythmia classification and transferred the knowledge learnt from arrhythmia to classify myocardial disease. The existing deep residual CNN architecture is retrained with ECG beats of myocardial infarction disease, which is proposed in the present work. We used the publically available kaggle dataset for the experimental evaluation of the proposed method. The present work improves the existing performance of the myocardial disease classification from 95% to 99%.

Accurate demand forecasting is necessary for choosing the right combination of compressors in a compressed air system for better energy saving. Also, prediction within the tolerance is a challenging task due to the stochastic nature of the demand. Therefore, this work proposes an efficient data driven approach for compressed air demand forecasting. We initially use the standard deep learning model Long Short Term Memory (LSTM) for prediction and forecasting. Further we use hybrid model in which we use LSTM to predict the required modes obtained from variational mode decomposition (VMD). We select the best model based on the prediction accuracy. The performance evaluation on the acquired datasets shows that the LSTM plus VMD model forecasts the demand within the specification.

The unprecedented availability of data in various fields reinforces the need for more comprehensive and advanced data-driven algorithms to deal with it. The data-driven techniques are able to handle the large volume of heterogeneous data and to extract more valuable information regarding the underlying system. This paper investigates the effectiveness of modern data-driven methods for data from non-linear systems. The effectiveness of dynamic mode decomposition (DMD) and its variants such as rSVD-DMD, randomized DMD (RDMD) and total DMD (TDMD) for dta from nonlinear system is investigated. The data considered for this study are cylindrical fluid flow and neuro recordings.

The world encountered a deadly disease by the beginning of 2020, known as the coronavirus disease (COVID-19). Among the different screening techniques available for COVID-19, chest radiography is an efficient method for disease detection. Whereas other disease detection techniques are time consuming, radiography requires less time to identify abnormalities caused by the disease in the lungs. In this study, one of the standard deep learning architectures, VGGNet, is modified for classifying chest X-ray images under four categories. The planned model uses images of four classes, namely COVID, bacterial, normal, and viral images. The performance matrices of the planned model are compared with five deep learning architectures, namely VGGNet, AlexNET, GoogLeNET, Inception-v4, and DenseNet-201.

This paper proposes and compares two nonlinear time-domain models of self-excited thermoacoustic oscillations of laminar premixed flames. Both models are hybrid formulations, where the flame and its immediate vicinity are resolved with a reactive flow simulation, while the acoustic field is modeled with a low-order model that is coupled to the reactive flow simulation. Firstly, a flame model based on the fully compressible Navier–Stokes equations is investigated. In this case, the flame simulation is coupled to the low-order model via the characteristic wave amplitudes at the inlet boundary. Secondly, the flame is resolved with a low Mach number reactive flow simulation. In order to include two-way thermoacoustic feedback, this flame model is coupled with an acoustic network model via the global heat release rate and the fluctuation of the axial velocity at a reference position upstream of the flame. A bifurcation analysis using the plenum length as bifurcation parameter is conducted. Both models exhibit complex nonlinear oscillations and are in good agreement with each other. Therefore, we conclude that the coupling of a linear acoustic model and a nonlinear flame model via reference velocity and global heat release rate is sufficient to accurately capture thermoacoustic oscillations of the configuration investigated. This implies that the most important nonlinearities can be attributed to hydrodynamic effects and flame kinematics. Furthermore, the study corroborates that premixed flames respond predominantly to fluctuations of the upstream flow velocity.