Department of Computer Science and Information Systems

Permanent URI for this collectionhttp://localhost:4000/handle/123456789/1928

Browse

Has files: NoSubject: search.filters.subject.Deep Learning

Search Results

Now showing 1 - 10 of 27
  • No Thumbnail Available
    Item
    A Survey on Recent Advances in Machine Learning Based Sleep Apnea Detection Systems
    (MDPI, 2021) Ramachandran, Anita
    Sleep apnea is a sleep disorder that affects a large population. This disorder can cause or augment the exposure to cardiovascular dysfunction, stroke, diabetes, and poor productivity. The polysomnography (PSG) test, which is the gold standard for sleep apnea detection, is expensive, inconvenient, and unavailable to the population at large. This calls for more friendly and accessible solutions for diagnosing sleep apnea. In this paper, we examine how sleep apnea is detected clinically, and how a combination of advances in embedded systems and machine learning can help make its diagnosis easier, more affordable, and accessible. We present the relevance of machine learning in sleep apnea detection, and a study of the recent advances in the aforementioned area. The review covers research based on machine learning, deep learning, and sensor fusion, and focuses on the following facets of sleep apnea detection: (i) type of sensors used for data collection, (ii) feature engineering approaches applied on the data (iii) classifiers used for sleep apnea detection/classification. We also analyze the challenges in the design of sleep apnea detection systems, based on the literature survey.
  • No Thumbnail Available
    Item
    Reputation-Based Reinforcement Algorithm for Motivation in Crowdsourcing Platform
    (Springer, 2019-07) Anand, Vijayalakshmi
    Crowdsourcing is a well-known model for solving tasks in several organizations in the recent times. While building, the crowdsourcing platform is simple, and its success depends on the amount of individuals taking part in it. We propose a brand new gamification methodology to draw folks to participate within the crowdsourcing platform. Reinforcement algorithm is employed in this gamification method to motivate the people. This reinforcement algorithm can direct a user to participate in some actions that yield maximum reward in a crowdsourcing platform. This gamification technique motivates user to participate in various activities in the crowdsourcing platform. The proposed algorithm is applied on a social media application that has been implemented for faculties to share their research and tutorial experience. We proved that participation of faculties in crowdsourcing platform improved after applying this gamification method.
  • No Thumbnail Available
    Item
    Deep Neural Networks for Securing IoT Enabled Vehicular Ad-Hoc Networks
    (IEEE, 2021) Alladi, Tejasvi; Chamola, Vinay
    Vehicular ad-hoc network (VANET) security has been an active area of research over the past decade. However, with the increasing adoption of the Internet of Things (IoT) in VANETs, the number of connected vehicles is set to grow exponentially over the next few years, which translates to a higher number of communication interfaces and a greater possibility of cybersecurity attacks. Along with these cybersecurity attacks, the instances of compromised vehicles sending faulty information about their positions and speeds also increase exponentially. Thus, there is a need to augment the existing security schemes with anomaly detection schemes which can differentiate normal vehicle data from malicious and faulty data. Since, the number of anomaly types can be many, deep neural networks would work best in this scenario. In this paper, we propose a deep neural network-based vehicle anomaly detection scheme. We use a sequence reconstruction approach to differentiate normal vehicle data from anomalous data. Numerical results show that we can correctly detect data corresponding to several anomaly types.
  • No Thumbnail Available
    Item
    Artificial Intelligence (AI)-Empowered Intrusion Detection Architecture for the Internet of Vehicles
    (IEEE, 2021-06) Alladi, Tejasvi; Chamola, Vinay
    Recent advances in the Internet of Things (IoT) and the adoption of IoT in vehicular networks have led to a new and promising paradigm called the Internet of Vehicles (IoV). However, the mode of communication in IoV being wireless in nature poses serious cybersecurity challenges. With many vehicles being connected in the IoV network, the vehicular data is set to explode. Traditional intrusion detection techniques may not be suitable in these scenarios with an extremely large amount of vehicular data being generated at an unprecedented rate and with various types of cybersecurity attacks being launched. Thus, there is a need for the development of advanced intrusion detection techniques capable of handling possible cyberattacks in these networks. Toward this end, we present an artificial intelligence (AI)-based intrusion detection architecture comprising Deep Learning Engines (DLEs) for identification and classification of the vehicular traffic in the IoV networks into potential cyberattack types. Also, taking into consideration the mobility of the vehicles and the realtime requirements of the IoV networks, these DLEs will be deployed on Multi-access Edge Computing (MEC) servers instead of running on the remote cloud. Extensive experimental results using popular evaluation metrics and average prediction time on a MEC testbed demonstrate the effectiveness of the proposed scheme.
  • No Thumbnail Available
    Item
    Securing the Internet of Vehicles: A Deep Learning-Based Classification Framework
    (IEEE, 2021-06) Alladi, Tejasvi; Chamola, Vinay
    Along with the various technological advancements, the next generation vehicular networks such as the Internet of Vehicles (IoV) also bring in various cybersecurity challenges. To effectively address these challenges, in addition to the existing authentication techniques, there is also a need for identification of the misbehaving entities in the network. This letter proposes a deep learning-based classification framework to identify potential misbehaving vehicles before the communication requests from the On Board Units (OBUs) of the vehicles can be entertained by the network infrastructure such as the Road Side Units (RSUs). The evaluated metrics demonstrate the performance of the proposed classification approaches.
  • No Thumbnail Available
    Item
    EraisNET: An Optical Flow based 3D ConvNET for Erasing Obstructions
    (IEEE, 2022) Narang, Pratik; Rajput, Amitesh Singh
    Images captured from behind a fence, window, or during rain generally face occlusions. Though prior works have addressed the problems of individually de-raining, reflection, and occlusion removal, a common approach that removes all the obstruction has found little attention in the literature. In this paper, we address the image occlusion problem by proposing a deep learning-based approach wherein the proposed method uses motion differences between two images and extracts important moving features from videos to separate the background and the obstruction. To accomplish this task, a novel 3D-convolution architecture is introduced, which is trained with synthetically blended videos. We have used learned layer-based CNN methods combined with dense-optical flow with generative networks for better output images. Moreover, a dataset for obstruction removal with sequences for reflection and fencing removal is proposed. The proposed approach is well experimented over a different variety of images and is found as a good candidate against state-of-the-art schemes.
  • No Thumbnail Available
    Item
    Privacy-preserving human action recognition as a remote cloud service using RGB-D sensors and deep CNN
    (Elsevier, 2020-08) Rajput, Amitesh Singh
    Cloud-based expert systems are highly emerging nowadays. However, the data owners and cloud service providers are not in the same trusted domain in practice. For the sake of data privacy, sensitive data usually has to be encrypted before outsourcing which makes effective cloud utilization a challenging task. Taking this concern into account, we propose a novel cloud-based approach to securely recognize human activities. A few schemes exist in the literature for secure recognition. However, they suffer from the problem of constrained data and are vulnerable to re-identification attack, where advanced deep learning models are used to predict an object’s identity. We address these problems by considering color and depth data, and securing them using position based superpixel transformation. The proposed transformation is designed by actively involving additional noise while resizing the underlying image. Due to this, a higher degree of obfuscation is achieved. Further, in spite of securing the complete video, we secure only four images, that is, one motion history image and three depth motion maps which are highly saving the data overhead. The recognition is performed using a four stream deep Convolutional Neural Network (CNN), where each stream is based on pre-trained MobileNet architecture. Experimental results show that the proposed approach is the best suitable candidate in “security-recognition accuracy (%)” trade-off relation among other image obfuscation as well as state-of-the-art schemes. Moreover, a number of security tests and analyses demonstrate robustness of the proposed approach.
  • No Thumbnail Available
    Item
    Multiclass Fake News Detection using Ensemble Machine Learning
    (IEEE, 2019) Narang, Pratik
    Over the past few years, fake news and its influence have become a growing cause of concern in terms of debate and public discussions. Due to the availability of the Internet, a lot of user-generated content is produced across the globe in a single day using various social media platforms. Nowadays, it has become very easy to create fake news and propagate it worldwide within a short period of time. Despite receiving significant attention in the research community, fake news detection did not improve significantly due to insufficient context-specific news data. Most of the researchers have analyzed the fake news problem as a binary classification problem, but many more prediction classes exist. In this research work, experiments have been conducted using a tree-based Ensemble Machine Learning framework (Gradient Boosting) with optimized parameters combining content and context level features for fake news detection. Recently, adaptive boosting methods for classification problems have been derived as gradient descent algorithms. This formulation justifies key elements and parameters in the methods, which are chosen to optimize a single common objective function. Experiments are conducted using a multi-class dataset (FNC) and various machine learning models are used for classification. Experimental results demonstrate the effectiveness of the ensemble framework compared to existing benchmark results. Using the Gradient Boosting algorithm (an ensemble machine learning framework), we achieved an accuracy of 86% for multi-class classification of fake news having four classes.
  • No Thumbnail Available
    Item
    FNDNet – A deep convolutional neural network for fake news detection
    (Elsevier, 2020-06) Narang, Pratik
    With the increasing popularity of social media and web-based forums, the distribution of fake news has become a major threat to various sectors and agencies. This has abated trust in the media, leaving readers in a state of perplexity. There exists an enormous assemblage of research on the theme of Artificial Intelligence (AI) strategies for fake news detection. In the past, much of the focus has been given on classifying online reviews and freely accessible online social networking-based posts. In this work, we propose a deep convolutional neural network (FNDNet) for fake news detection. Instead of relying on hand-crafted features, our model (FNDNet) is designed to automatically learn the discriminatory features for fake news classification through multiple hidden layers built in the deep neural network. We create a deep Convolutional Neural Network (CNN) to extract several features at each layer. We compare the performance of the proposed approach with several baseline models. Benchmarked datasets were used to train and test the model, and the proposed model achieved state-of-the-art results with an accuracy of 98.36% on the test data. Various performance evaluation parameters such as Wilcoxon, false positive, true negative, precision, recall, F1, and accuracy, etc. were used to validate the results. These results demonstrate significant improvements in the area of fake news detection as compared to existing state-of-the-art results and affirm the potential of our approach for classifying fake news on social media. This research will assist researchers in broadening the understanding of the applicability of CNN-based deep models for fake news detection.
  • No Thumbnail Available
    Item
    ReViewNet: A Fast and Resource Optimized Network for Enabling Safe Autonomous Driving in Hazy Weather Conditions
    (IEEE, 2020) Narang, Pratik; Chamola, Vinay
    Adverse weather conditions such as fog, haze, snow, mist and glare create visibility problems for applications of autonomous vehicles. To ensure safe and smooth operations in frequent bad weather scenarios, image dehazing is crucial to any vehicular motion and navigation task on road or air. Moreover, the commonly deployed mobile systems are resource constrained in nature. Therefore, it is important to ensure memory, compute and run-time efficiency of dehazing algorithms. In this manuscript we propose ReViewNet, a fast, lightweight and robust dehazing system suitable for autonomous vehicles. The network uses components like spatial feature pooling, quadruple color-cue, multi-look architecture and multi-weighted loss to effectively dehaze images captured by cameras of autonomous vehicles. The effectiveness of the proposed model is analyzed by exhaustive quantitative evaluation on five benchmark datasets demonstrating its supremacy over other existing state-of-the-art methods. Further, a component-wise ablation and loss weight ratio analysis demonstrates the contribution of each and every component of the network. We also show the qualitative analysis with special use cases and visual responses on distinctive vehicular vision instances, establishing the effectiveness of the proposed method in numerous hazy weather conditions for autonomous vehicular applications.