Weather prediction is an important aspect of modern society, with implications for everything from agriculture to disaster response. However, accurately predicting the weather remains a challenging task due to the complexity of atmospheric processes and the limitations of existing technologies. In recent years, advancements in futuristic technologies such as artificial intelligence and machine learning, high-resolution weather models, climate modeling, remote sensing, and citizen science have offered new opportunities to improve weather prediction. These technologies have the potential to provide more detailed and accurate data, identify patterns and relationships in large data sets, and improve communication and education strategies. Continued research and development in these areas may lead to more accurate and timely weather predictions, ultimately improving public safety and quality of life.
The advent of deep learning has revolutionized the technology industry and has made Deep Neural Networks (DNNs) the powerhouse of many modern day software applications. Well-trained DNNs are able to perform complex tasks such as speech recognition, object detection and image classification with high precision and accuracy. However, the task of training such complex networks at times requires enormous amount of computational resources for which the task is often outsourced to third parties. Recent work suggests that outsourcing the training task can act as a favourable gateway for a malicious trainer to induce a backdoor in the model, which when triggered can force the model into performing in a predefined way which has been set up by the malicious trainer. This paper starts by giving an overview on how such attacks are induced and consequently discusses and provides experimental proof on the various strategies which can be used to neutralise such attacks. We use the l1 and l2 norm to identify weights which are susceptible to be poisoned and prune them away by setting their value to zero. We further inspect the efficiency of layer wise and global pruning. We infer from our experiments that fine-tuning the model for a few epochs after the fine-pruning stage has been completed helps the model to regain its lost accuracy and provides better test time accuracy. During this study, we understand that performing fine-pruning in the later layers is more effective. By pruning the last layer along with fine-tuning the model after fine-pruning has been completed, we achieve 99.96% and 86.05% accuracy for the clean validation and test dataset respectively and consequently witness a drop in attach success rate from 99 to 0 and near 0 in some cases.
The prevalence of picture tampering and alteration has grown significantly in recent years across several industries. The objective of this research work is to identify and draw attention to violations committed when manipulating digital photographs nowadays. The system identifies areas of the image that are likely to be altered using a combination of feature extraction, image segmentation, and machine learning approaches. Its effectiveness is then tested using a publicly accessible dataset of altered photos.
Due to certain unique qualities and capabilities, Blockchain is a highly beneficial approach that may be used to safely manage various gadgets in a smart city. It has several uses, particularly in dispersed situations where elements such as wireless sensor nodes must be confident of the server's legitimacy. Because modern blockchain solutions that handle post-quantum problems have not been developed, we explore a blockchain in the quantum-resistant cryptography context and strive to find how it might withstand quantum computing assaults throughout this work. Furthermore, the newly born Proof of Stake (PoS) provides faster and cheaper transactions, but its security is not proven in comparison to its forefather Proof of Work (PoW). As a result, a new quantum-resistant proof of stake (Quantum-Resistant PoS) agreement technique for smart city application services has been developed. The proposed model has the ability to protect a distributed ledger system against a quantum assault and also provides scalability and inexpensive transactions. Additionally, user-based post-quantum authentication is included in the transaction process to create a simple payment verification node. Following that, we present a detailed rundown of how to execute a post-quantum simple payment verification and transaction on a blockchain. Thus, our study will contribute to imminent quantum-resistant blockchain exploration along with the design or structure of potentially distributed ledger technology-based ubiquitous computing.
Cybersecurity has emerged as one of the most crucial facets of the Internet of Things (IoT) due to the increased possibility of cyberattacks. IoT cybersecurity aims to lower cybersecurity risk for businesses and users by safeguarding IoT resources and privacy. IoT security management could be improved by using new cybersecurity technology and solutions. However, efficient IoT cyber risk management frameworks are lacking for organizations. This paper presents the most significant proposals of cybersecurity in IoT by describing their objectives, working principles, applications, features, and approaches referred. The paper also presents cybersecurity in IoT using five optimization algorithms such as particle swarm optimization, ant colony optimizations, an artificial bee colony, genetic algorithm, and AdaBoost algorithm. The paper concludes with research gaps and limitations, which urge for further study for subsequent research work.
Does an iOS app preserve a user's privacy? With the presence of 1.76 million apps on the App Store, Apple Inc. hosts the second largest online store than its counterfeit Android. iOS apps have gained enormous popularity amongst users since they are used for social interaction, banking, navigation, and sharing personal experiences, making it usual for users to share their information via these apps at the stake of their privacy. Preserving user privacy has become an important concern and measuring privacy leaks by iOS apps is a challenging task. It is very difficult for users to preserve their privacy without a practical and efficient way to quantify and evaluate privacy. In this paper, we have proposed a framework iPDS, that work in two parallels. First, it computes the privacy disclosure score of 723 iOS apps. The privacy disclosure score (PDS) is computed using two metrics sensitivity and visibility across three privacy settings, link, not link and track for four categories of iOS apps. The categories are Education, Finance, Games, and Entertainment. Second, machine learning classifiers have been applied for classification of malicious with non-malicious apps and to check the permissions settings that are suitable for classification. Our experimental results indicate that the track permission settings along with Naïve Bayes algorithm are found to be best for classification that has attained 93.4%, 93.8%, and 97.5% for three categories Education, Entertainment & Games.
A polyp is an abnormal growth that can occur in the colon and can be a precursor to colorectal cancer. Colorectal cancer can be diagnosed and treated earlier when polyps are detected and segmented accurately in medical images. The use of deep learning models for medical image segmentation has shown promising results in recent years. This paper presents a polyp segmentation using the modified version of Deeplabv3+ with an attention mechanism called AB-DeepLabv3+. An attention mechanism enhances the discriminatory power of the network by highlighting informative regions of the input image. In the evaluation of the AB-DeepLabv3+ on the publicly available Kvasir-SEG dataset, it achieved state-of-the-art performance with an overall Dice coefficient of 0.98 and an intersection over union of 0.96. The results demonstrated that the AB-DeepLabv3+ provides accurate and efficient polyp segmentation in colonoscopy images.
There are many Internet of Things (IoT) applications which demand an infrastructure that requires significantly less energy and supports longer operation time. Such an infrastructure is provided by Low-Power and Lossy Networks (LLN) that run on resource-constrained devices. Achieving an energy-efficient routing on such networks is a significant challenge which is addressed by the Routing Protocol for Low-power and Lossy Networks (RPL). The specifications of RPL are depicted in RFC 6550 by IETF. Although RPL gives many benefits to LLN, at the same time, it also attracts many security issues. One such security issue is the Dropping Destination Advertisement Object (DDAO) attack. In RPL, a client node uses Destination Advertisement Object (DAO) control messages to pass the destination information to the root node. An attacker may exploit the DAO passing mechanism of RPL to deliberately drop DAO received from child nodes and send fake Destination Advertisement Object-Acknowledgement (DAO-ACK) messages to perform a DDAO attack. The existing research work provides only a detection algorithm to detect the DDAO attack. Therefore, in this research paper, we proposed a defense technique, i.e., lightweight Acknowledgement (ACK) authentication, based on a challenge-response strategy to mitigate the attack. For this, we have used a Prime Sequence Code Matrix (PSCM) and a modified version of the DAO and DAO-ACK control packet of the standard RPL. Our proposed technique reduces the impact of attack and restores network performance significantly.
Detecting the Arrhythmia correctly is an important task in cardiovascular disease detection. Identifying the correct type of arrhythmia provides a fast and effective way for diagnosis. In this paper a neural network based (ANN & CNN) and fuzzy logic based classifiers (ANFIS) are presented and their performance is evaluated using performance evaluation parameters. For this study, MITBIH and PTB diagnostic database are used which provided the total sample size of 1,23,998 ECG waveforms. For classification purpose, features are extracted such as Skewness, Kurtosis, Mean, Standard Deviation, Shannon Entropy, and Energy. In this study, it has been observed that ANFIS based classifier performs better than neural network based classifier.
Abstract The problem of road accidents is a serious global issue that results in countless deaths and injuries every year. To reduce the number of fatalities, it is essential to detect accidents promptly and provide immediate assistance to the injured. Numerous techniques have been proposed to detect road accidents and alert hospitals, but there is still room for improvement. This project aims to develop a highly accurate image-based system for assessing road accidents in real-time. The system will gather information from surveillance cameras and use machine learning algorithms such as Convolutional Neural Networks (CNN) to process the data. The proposed system records video streams, analyses the input, and generates signals that are sent to emergency services such as ambulances or fire departments in real-time. The key advantage of this system is that it uses CCTV cameras to detect injuries and alert emergency services quickly, potentially saving lives. By leveraging advanced machine learning techniques, the system can accurately identify and assess the severity of accidents, enabling first responders to arrive on the scene with the necessary equipment to provide immediate medical assistance. Overall, this project represents a significant step forward in the development of automated systems for detecting and assessing road accidents. The proposed system has the potential to save lives by improving the speed and accuracy of emergency response times, ultimately contributing to a safer and more efficient transportation system.
The zero-trust security model is a modern alternative to traditional perimeter network security and has been gaining popularity over the last several years. Traditional perimeter security suffers the problem of single-point failure and is compelling enterprises to move away from perimeter security. The paradigm of zero trust addresses this vulnerability through the handling of every host as though they are faced with the Internet. As the number of organizations moving towards working from home increases and continues to be the new normal, the definition of trust also needs to be evolved. This paper discusses the zero trust protection model architecture, its accomplishment measures and benefits relative to other network models. We propose a multi-layer zero trust architecture which includes three cascaded layers, four enablers and five security attributes. We also suggest a six-stage implementation framework along with the tools aiding the stage. We summarize the paper by reinforcing the advantages of the zero-trust paradigm and listing down future research directions. This paper can contribute to the knowledge base by directing organizations towards a successful paradigm shift to the zero-trust environment.
This paper presents a bibliometric analysis of the literature on blockchain technology's application in human resource management. The study aims to map the current state of research on this topic and identify major themes and trends in the literature. We analyzed a total of 124 articles from 32 countries, authored by 238 individuals and 82 institutions from the Scopus database. Our findings indicate that research on blockchain in human resource management is a growing area of interest, with a focus on topics such as recruitment, talent management, and employee data management. We also found that most of the research has been conducted in North America (42.7%) and Europe (34.7%), with fewer contributions from Asia (14.5%) and other regions. Our study contributes to the literature by providing a comprehensive overview of the current knowledge on blockchain in human resource management and identifying research gaps and future directions.
The novel coronavirus disease or COVID-19 is a highly infectious disease and it has infected approximately 150 million people across the world and 20 million people in India. The governments, the hospitals, and even the people themselves are working hard to control the pandemic since March 2020. Various technologies have been used to control the pandemic and Cloud Computing has come to light to be an appropriate solution as it provides better availability, less costs, and security. This research paper presents an efficient solution for monitoring and controlling the viral disease using the Cloud platform. The proposed Cloud architecture predicts and classifies the infected and non-infected people and thereafter works on monitoring the patients based on the severity level of their disease. The paper also focuses on controlling the disease using contact tracing of infected patients using Social Network Analysis (SNA). A dataset containing contract tracing information for the city of Bangalore was used to test on the proposed architecture. The system outperforms by providing an accuracy of 92% and also resource utilization of 81% was observed on the Amazon EC2 cloud platform. The paper's core idea is the usage of SNA graphs for contact tracing and hence, calculating Outbreak Role Index that indicates how much an infected person has spread the disease among his or her contacts. The architecture proposed can be very beneficial for the government and healthcare departments, to effectively analyze and prevent the COVID-19 outbreak.
In recent years, virtualization has become popular with the usage of emerging technologies in different domains. With the increasing number of attacking incidents in recent times, virtualization security has become one of the primary focus of research. The traditional network attack detection systems are inefficient enough to detect attacks over the virtual network. In this paper, we have proposed a virtual network security framework, called VNSecure that detects malicious network activities by analyzing virtual machine (VM) traffic profile. VNsecure operates on a hypervisor layer and has access to both underlying hardware and guest operating system. It does VM-level activity analysis from the privileged domain of the hypervisor, serving as the main line of defense against intrusions at the virtual network level. Initially, VM traffic validation is performed to detect spoofing attacks by analyzing the traffic captured at the backend driver of the virtual network interface of the monitored VM. To perform detailed behavior analysis, a deep learning approach is used to learn and detect VM-specific network attacks. On detection of malicious traffic, an alert is raised to the administrator. VNSecure then carries out essential mitigation to lower the risk and store the occurrence of malicious packets in its database. One public dataset and one self-generated attack dataset have been used to validate all the modules of the proposed framework and the results seem to be promising.
Our population is increasing day by day, and at the same time, the response to climate change has put enormous pressure on the agricultural sector to increase productivity and food production. The agricultural land is gradually reducing in most of the country and it is nearly impossible to increase it back again. Agricultural automation is the only option for precision farming in today’s era and is also the demand of today’s time. Artificial Intelligence (AI) have started contributing and capitalizing on precision farming and all industries long back. The use of digital technologies has revolutionized agriculture, which is helpful for precision farming by providing smart interfaces that can guide small farmers for new crops. Profitable crops and provide solutions for their crops-related queries. The important application of AI is to achieve a better yield, as well as to increase the quality of the crop, detection of diseases, weeds control, pest detection, an application of fertilizer at the right time, greenhouse, cultivation, crop health monitoring etc. These aspects have been discussed in this article. The main objective of this paper is how agriculture is being operated with digital technology in the field of agriculture. Through this paper, those researches have been observed, and the major applications that have been made so far in the field of agricultural science are to be identified with the help of AI.
The imbalanced dataset’s existing classification methods have low prediction accuracy for the minority class because of the little information present. Using over- and under-sampling techniques, we can improve the minority’s ability to forecast outcomes. However, the minority class’s accuracy of prediction is negatively impacted by the two methods due to the loss of vital information or the addition of irrelevant details for classification. SVM kernels have great abilities to handle asymmetric data, but when we need to use SVM kernels alone or as part of the ensemble technique for an unbalanced dataset, we don’t have a strong reason to choose which kernel to use, and also how a particular kernel will act depends a lot on the data set. In this paper, we present a framework in which several kernel SVM (Linear, Polynomial, Sigmoid, RBF) classifiers were utilized as the base learners and one of the kernels (say RBF kernel) as meta learner using the Stacking Ensembles technique, which shows that stacked generalization of SVM kernels gives similar results as best performing kernel for an imbalanced dataset of software change proneness, using AUC, ROC, MCC, and BAS as an evaluation matrix.
The cloud environment has become an essential platform due to its computing abilities and is being used in various fields and sectors all around the globe. Users from all over the globe use this computing platform to process their challenging tasks. The cloud computes these tasks on its Virtual Machines (VM) using the appropriate resource scheduling algorithms. While a particular task is being computed, there is always a chance that the cloud suffers damages due to the dynamically generated faults of the task. The cloud also needs better performance with proper resource scheduling, leading to increased costs. To focus on these problems and provide an intelligence mechanism to the cloud, an algorithm named Reinforcement Learning – First Come, First Serve (RL – FCFS) has been designed and implemented by combining the Reinforcement Learning (RL) technique with the existing resource scheduling algorithm First Come First Serve (FCFS) to handle the dynamic faults and provide better cost by improving the resource scheduling at its end. This RL – FCFS algorithm provides a fault-tolerance mechanism at the cloud's end by computing 55.5 % of tasks aggregately compared to an aggregate of 11.1 % for the FCFS. Also, it aggregately improves the cost by 18.50 % across all scenarios. With the RL – FCFS algorithm, the cloud will be in a learning phase at the beginning. With RL rewards and feedback, the cloud will adapt and begin to handle these dynamic faults over time and improve its resource scheduling process, ultimately providing the best Quality of Service (QoS).
Over the last few years, Hybrid Electric Vehicles (HEVs) have become increasingly popular due to their potential to simplify fuel consumption and greenhouse gas emissions. The energy management of HEVs is a critical task that involves controlling the power split between the Internal Combustion Engine (ICE) and electric motor based on the vehicle’s state and driving conditions. Traditional rule-based strategies for HEV energy management may not be able to adapt to varying driving conditions or optimize the vehicle’s performance in real-time. To address this, researchers have explored the potential of advanced machine learning techniques, such as Deep Reinforcement Learning (DRL), as a more effective approach for HEV energy management. DRL is a subfield of machine learning that combines deep neural networks with reinforcement learning to learn an optimal control policy. Among various DRL algorithms, Deep Dyna-Q learning is a hybrid approach that combines model-based and model-free learning. Our paper introduces an innovative strategy for energy management for HEVs using Deep Dyna-Q learning that optimizes the power split in real-time based on the vehicle’s state and driving conditions. We evaluate the proposed strategy on two driving cycles and compare it with Deep Q-Learning (DQL). The findings indicate that the energy management approach presented in this paper surpasses DQL in terms of vehicle performance and fuel efficiency for both driving cycles.
Social network analysis has gained popularity as a field of study in recent years, with link prediction being a critical task. Link prediction involves inferring missing or future connections between nodes in a network. Machine learning has become a widely used technique for link prediction in social networks due to its ability to detect complex patterns in network data. This paper proposes a novel machine learning-based approach for link prediction in social networks. The approach incorporates global similarity measures, such as the Normalized Katz Index and Normalized Path distance, and local similarity measures, such as the Jaccard Coefficients and Adamic Adar, to enhance the prediction accuracy. This research paper claims that using only global or local similarity measures is insufficient for accurate link prediction; however, combining these two measures improves the accuracy significantly. The proposed approach was evaluated using three real-world social network datasets - PowerGrid, Caltech, and Circuit - and three models were employed, namely Decision Tree, Random Forest, and Support Vector Machine. The results indicate an accuracy rate of over 82 %, which surpasses the performance of several state-of-the-art link prediction techniques.
Bug triaging is a crucial activity in software development that involves identifying and prioritizing bugs for fixing. But recommending cold bugs to an appropriate developer is a more challenging problem. In this work, an empirical assessment of the effectiveness of two reinforcement learning algorithms, Multiarmed bandits and Contextual bandits, in the context of triaging cold-start bugs, is investigated. Five publicly available open-source bug-triaging datasets have been used to evaluate the performance of the algorithms using two standard evaluation metrics i.e., rewards and average rewards. Evaluations are done of two MAB algorithms- ɛ-Greedy and UCB, and LinUCB as a Contextual MAB algorithm. Our results showed that both Multiarmed bandits and Contextual bandits are effective in triaging cold start bugs but in different settings. Contextual bandits outperformed Multiarmed bandits in terms of rewards and average reward in different simulation settings. Our results demonstrate that contextual MAB algorithms outperform traditional MAB algorithms in accurately recommending cold bugs to an appropriate developer.
