Abstract

Deep learning techniques have shown significant impact in enhancing security across various domains by leveraging artificial neural networks models. When applied to cloud computing security, deep learning offers cost-effective solutions by automating threat detection, reducing manual monitoring, and improving overall security effectiveness. Deep learning models using neural networks play crucial role in tasks like intrusion detection, malware detection, anomaly detection, and log analysis. Integration of deep learning into cloud security requires careful evaluation of existing systems, defining objectives, dataset selection and preparation, model tuning, and eventual modifications for compatibility. Furthermore, implementing deep learning techniques in cloud security entails considering factors such as computational resources, data collection and preparation costs, model development, integration efforts, and ongoing monitoring and maintenance. This paper proposes a feed-forward propagation Artificial Neural Network (ANN) model in cloud security and investigates the key steps for integrating such models into cloud security strategies. Considering that the effectiveness of the ANN model depends on factors such as training data quality, network architecture, and weight adjustment algorithms, the study utilizes a dataset from Kaggle.com for validation and demonstrates steps involved in training and evaluation of the ANN model.

Introduction

Deep learning (DL) techniques have emerged to become effective tools for enhancing security in various domains. These techniques leverage the capabilities of artificial neural networks to learn and identify patterns from vast amounts of data, enabling more robust and efficient security solutions [1][2]. When applied to cloud computing security, deep learning techniques offer several benefits having direct implications in the cost-effectiveness of the products. A great benefit of implementation of DL in security of cloud is seen in automated threat detection. DL algorithms can analyze large volumes of data, such as network traffic logs, system logs, and user behavior, to detect anomalies and potential security threats automatically [2], [3]. This together with other utilizations as a result of DL, reduces the need for manual monitoring and analysis, what leads to faster identification of security incidents, timely response, mitigation, and reduced costs in time and resources. Considering this, through automation of various security tasks, DL minimizes the risk of human error, which can lead to security breaches and associated costs.

Automated systems powered by deep learning can perform tasks consistently and accurately, enhancing overall security effectiveness [4], [5]. DL models learn patterns and relationships in data, allowing for more accurate threat detection and classification. Through detecting patterns and anomalies which could go unnoticed by traditional rulebased systems, DL implementation makes it possible for organizations to optimize resource allocation and minimize unnecessary costs.

Conclusion

The integration of deep learning techniques into cloud computing security offers numerous benefits. By analyzing large volumes of data, deep learning algorithms can detect security threats in real-time and minimize various risks. However, the implementation of deep learning in cloud security requires careful evaluation of costs and resources. Organizations struggle when having to adjust the solution to the computational resources, data acquisition and preparation costs, personnel expertise, and ongoing monitoring and maintenance.

The integration of deep learning techniques in cloud security involves few steps for a successful outcome. These include assessing the strengths and weaknesses of the existing system, identifying specific tasks that deep learning can address, selecting suitable models and algorithms, training the models with labeled data, and ensuring compatibility and integration with existing components. Continuous monitoring and feedback are also crucial stages for maintaining the efficiency and effectiveness of the integrated system.

Abstract

Introduction

Material and methods

Results and discussion

Conclusions

Acknowledgments

Supplementary materials

Reference

Abstract

     Air pollution is a major obstacle to future sustainability, and traffic pollution has become a large drag on the sustainable developments of future metropolises. Here, combined with the large volume of real-time monitoring data, we propose a deep learning model, iDeepAir, to predict surface-level PM2.5 concentration in Shanghai megacity and link with MEIC emission inventory creatively to decipher urban traffic impacts on air quality. Our model exhibits high-fidelity in reproducing pollutant concentrations and reduces the MAE from 25.355 μg/m3 to 12.283 μg/m3 compared with other models. And identifies the ranking of major factors, local meteorological conditions have become a nonnegligible factor. Layer-wise relevance propagation (LRP) is used here to enhance the interpretability of the model and we visualize and analyze the reasons for the different correlation between traffic density and PM2.5 concentration in various regions of Shanghai. Meanwhile, As the strict and effective industrial emission reduction measurements implementing in China, the contribution of urban traffic to PM2.5 formation calculated by combining MEIC emission inventory and LRP is gradually increasing from 18.03% in 2011 to 24.37% in 2017 in Shanghai, and the impact of traffic emissions would be ever-prominent in 2030 according to our prediction.

Introduction

     Air pollution is a large obstacle to the world’s future sustainable developments, and millions of people die from air pollution-related diseases every year around the world (Zheng et al., 2017). This is seriously severe in some developing countries like China (Lelieveld et al., 2015), which has the highest country-level values globally for the population-weighted annual average concentration of PM2.5 (Tichenor and Sridhar, 2019; Zhang et al., 2012) and has been a major public health concern in recent years (Li et al., 2019a) . Shanghai, one of the most developed and populous cities in China, has suffered severe increasing haze episodes mostly attributed to the severe particle pollution especially fine particles (particles ≤2.5μm in aerodynamic diameter; PM2.5)(Han et al., 2020) since 1990s with the rapid urbanization and industrialization (Wang et al., 2015) . Under these circumstances, air pollution-related diseases have emerged gradually, such as respiratory diseases in the elderly and preterm birth and low birth weight for birth when maternal exposure to PM2.5 in Shanghai(Li et al., 2019a; Liu et al., 2017) .

Conclusions

     In this article, we propose a novel method to quantify the influence of anthropogenic emissions on PM2.5 concentration with a novel and traceable deep learning model. Our experiment results indicate that the proposed model could achieve better fitting and prediction performances than the LSTM, GBRT, Seq2Seq, DA-RNN model and other deep learning models. Furthermore, we output the contributions of input parameter to the prediction in LRP method and visualize and analyze the spatial correlation between traffic flow and PM2.5.

     In addition, we discover that transportation emissions will play the most dominate role in future urban air pollution, and how to reduce traffic emissions are an unavoidable issue on achieving sustainable developments in modern cities. Meanwhile, to some extent, new energy vehicles can be considered as an effective way to reduce traffic emissions. However, the current promotion policies and efforts are far from enough. To further improve urban air quality, we need some more effective and powerful measurements in response to the rapid growth of urban traffic emissions.

Abstract

Electroencephalography (EEG), which tracks the brain waves that contain the brain’s neural activity, plays an essential role in detecting human motion and treating neurological diseases. In the Artificial Intelligence (AI) era, deep learning algorithms are widely used in human action recognition and classification. Various convolutional neural networks that process this signal are also being born. This paper provides a detailed survey of the application of deep learning to EEG signals and outlines the research process when classifying EEG signals. At the same time, this paper reviews the relevant research on the classification of human action EEG signals in recent years. Human motion signals usually use different deep learning algorithms and convolutional neural network architectures in the EEG signal analysis task. This article will discuss the advantages and challenges of each method in other studies. Finally, the paper discusses future directions for deep learning-based EEG signal classification.

Abstract

     Concrete is one of the most widely used materials in various civil engineering applications. Its global production rate is increasing to meet demand. Mechanical properties of concrete are among important parameters in designing and evaluating its performance. Over the past few decades, machine learning has been used to model real-world problems. Machine learning, as a branch of artificial intelligence, is gaining popularity in many scientific fields such as robotics, statistics, bioinformatics, computer science, and construction materials. Machine learning has many advantages over statistical and experimental models, such as optimal accuracy, high-performance speed, responsiveness in complex environments, and economic cost-effectiveness. Recently, more researchers are looking into deep learning, which is a group of machine learning algorithms, as a powerful method in matters of diagnosis and classification. Hence, this paper provides a review of successful ML and DL model applications to predict concrete mechanical properties. Several modeling algorithms were reviewed highlighting their applications, performance, current knowledge gaps, and suggestions for future research. This paper will assist construction material engineers and researchers in selecting suitable and accurate techniques that fit their applications.

Introduction

     Concrete is the most widely used building material worldwide. With population growth and urbanization, the demand for concrete is expected to reach 18 billion by 2050 [[1], [2], [3]]. In order to improve the design of concrete structures, it is necessary to gain a better understanding of concrete performance, relying on accurate evaluation of its mechanical properties. Among the various properties of concrete, compressive strength has been considered a direct indicator of performance. It is directly related to the safety and performance of the structure throughout its life cycle [1,4,5].

     Concrete is a complex system of combinations of different components (coarse and fine aggregates, water, cement, and additional mixtures) that are randomly distributed throughout the concrete matrix [[6], [7], [8], [9]]. This heterogeneous feature makes it difficult to accurately predict certain mechanical properties, especially compressive strength [[10], [11], [12]]. The most direct way to evaluate the compressive strength of concrete is through physical tests performed on specimens cured to the desired age [13,14]. Such a method for evaluating the compressive strength needs time while being affected by other factors related to specimen fabrications and test operators. Moreover, the test tends to damage the specimens. Empirical regression methods were therefore proposed to predict compressive strength [[15], [16], [17]], but the disadvantage of this method is the non-linear relationship between the concrete mixture and the concrete’s compressive strength. This prevents an accurate regression expression. Numerical simulation is another method that can predict the behavior of concrete. However, a good prediction of concrete behavior is not easy to achieve due to the non-linearity and randomness [[18], [19], [20]].

Conclusions and recommendations

     The present study investigated the performance of different ML/DL models to predict the mechanical properties of concrete. Among the wide range of ML/DL models, support vector machines, decision trees, evolutionary algorithms, and artificial neural networks were examined due to the popularity of these models as well as their frequent use in the field of civil engineering. Therefore, the performance of these models in various studies to evaluate the compressive strength, tensile strength, shear strength, flexural strength, and elastic modulus was evaluated. A comparison of experimental models with ML/DL methods shows that ML/DL models with better updating capability and the ability to analyze large datasets perform better. On the other hand, statistical models are not good for predicting complex structures due to their high cost and time-consuming nature. According to ML/DL methods, selecting the appropriate model for predicting the target strength of concrete should be made by considering different criteria. A review of various studies shows that the relationship between concrete components and mechanical strength is influential in choosing the forecasting model. Therefore, models that can respond in nonlinear space should be used if the relationship is nonlinear. In these cases, SVM and ANN models can be used for their acceptable performance in a non-linear environment with fewer errors. To achieve more accurate results, optimization of these models with metaheuristic algorithms can also be used. But if we need the transparency of the model and the explicit mathematical formula between input and output, we can use decision tree models and evolutionary algorithms. On the other hand, the use of ensemble models to optimize these models, although it results in higher accuracy, increases computation time and model complexity. The combined SVM and ANN models, although they increase the computation time, have accurate results in the face of extensive data. Therefore, based on the study, the best option for predicting the strength of concrete in terms of model accuracy and model implementation is the combined use of SVM and ANN combined models.

Abstract

     The automotive industry is undergoing a profound digital transformation to create autonomous vehicles. Vehicle-to-Everything (V2X) communications enable the provisioning of transportation use cases for road traffic and safety management. At the same time, during the past decade, Artificial Intelligence (AI) and Machine Learning (ML) have been in the spotlight because of their outstanding performance in various domains, including natural language processing, and computer vision. Considering also current standardization efforts, towards incorporating AI and ML as integral sub-systems of beyond 5G and 6G networks, these technologies are considered very promising to optimize user, control, and management network functions, but also to support road safety and even entertainment applications. This survey systematically reviews existing research at the intersection of AI/ML and V2X communications, focusing on handover management, proactive caching, physical and computation resources allocation, beam selection optimization, packet routing, and QoS prediction in vehicular environments. We extract the underlying AI/ML techniques, the training features, their architecture and discuss several aspects regarding the intricacies of vehicular environments and ML. These aspects include time complexity of the algorithms, quality of real-world vehicle traces, suitability of AI/ML techniques in relevance to the designated network operation and the underlying automotive use case, as well as velocity and positioning accuracy requirements towards the creation of more realistic and representative synthetic data.

Introduction

     During the past years, car manufacturers have introduced driver assistance systems to their models, coupled with onboard intelligence, leading to a higher perception of their surroundings. This enables the possibility to achieve different levels of autonomous driving. Autonomous driving is considered critical in improving car safety, eliminating accidents due to human error, reducing traffic congestion, and improving passenger comfort. The Society of Automotive Engineers (SAE) has defined six driving automation levels, ranging from no automation to full automation [1]. Communications among vehicles, infrastructure and road users, collectively defined as Vehicle-to-Everything (V2X), are essential in realizing safety and non-safety-related applications, such as autonomous driving, car platooning, information sharing among vehicles and high data-rate infotainment.

     V2X may refer to Vehicle-to-Vehicle (V2V), Vehicle-to-Infrastructure (V2I), Vehicle-to-Network (V2N) or Vehicle-to-Pedestrian (V2P) communication. V2P refers to the communication among vehicles and pedestrians, cyclists or motorized two-wheeler operators, collectively called Vulnerable Road Users (VRUs) [2]. There are two leading technologies in V2X: i) the Cellular-V2X (C-V2X) based on cellular 4G/LTE [3] and 5G networks [4] and ii) the Dedicated Short-Range Communication (DSRC) [5] based on IEEE 802.11p [6]. These solutions will complement the sensor/camera/radar information and intelligently connect the car to its surroundings and the network.

Conclusion

     This survey presented recent advances in AI/ML applications in V2X communications. We classified the related literature in handover management, beam allocation, caching, radio network allocation, computations resources management, routing and QoS prediction considering only vehicular environments. For each category, we surveyed the ML technique, the training features, architecture, and optimization objectives, and extracted results and observations concerning time complexity, performance and suitability of learning techniques according to the designated problem

     g techniques according to the designated problem. Based on the surveyed publications, there is no “one size fits all” solution. Depending on the problem, different tasks require different formulations, including AI/ML algorithm selection, optimization objectives, architecture, and training features. Each family of AI/ML algorithms comes with their own advantages and disadvantages. The problem formulation must also consider the requirements of the underlying use case, which is largely affected by the training and response times of the selected AI/ML model. In addition, it is vital to explore approaches that reduce these times, while preserving the robustness of the AI/ML algorithm. AI/ML in V2X has already shown potential in optimizing network operations, but there are still open issues that need to be addressed due to the intricacies of both AI/ML and the highly dynamic vehicular networks. Based on the surveyed papers, vehicular networks, empowered by AI/ML and V2X communications as cooperative technologies, can be transformed into autonomous networks with self-configuration/optimization/healing capabilities.

Abstract

     A fundamental expectation of the stakeholders from the Industrial Internet of Things (IIoT) is its trustworthiness and sustainability to avoid the loss of human lives in performing a critical task. A trustworthy IIoT-enabled network encompasses fundamental security characteristics, such as trust, privacy, security, reliability, resilience, and safety. The traditional security mechanisms and procedures are insufficient to protect these networks owing to protocol differences, limited update options, and older adaptations of the security mechanisms. As a result, these networks require novel approaches to increase trust-level and enhance security and privacy mechanisms. Therefore, in this article, we propose a novel approach to improve the trustworthiness of IIoT-enabled networks. We propose an accurate and reliable supervisory control and data acquisition (SCADA) network-based cyberattack detection in these networks. The proposed scheme combines the deep-learning-based pyramidal recurrent units (PRU) and decision tree (DT) with SCADA-based IIoT networks. We also use an ensemble-learning method to detect cyberattacks in SCADA-based IIoT networks. The nonlinear learning ability of PRU and the ensemble DT address the sensitivity of irrelevant features, allowing high detection rates. The proposed scheme is evaluated on 15 datasets generated from SCADA-based networks. The experimental results show that the proposed scheme outperforms traditional methods and machine learning-based detection approaches. The proposed scheme improves the security and associated measure of trustworthiness in IIoT-enabled networks.

Introduction

     THE Industrial Internet of Things (IIoT) is a pervasive network that connects a diverse set of smart appliances in the industrial environment to deliver various intelligent services. In IIoT networks, a significant amount of industrial control systems (ICSs) premised on supervisory control and data acquisition (SCADA) are linked to the corporate network through the Internet [1]. Typically, these SCADA-based IIoT networks consist of a large number of field devices [2], for instance, intelligent electronic devices, sensors, and actuators, connected to an enterprise network via heterogeneous communications [3]. This integration provides the industrial networks and systems with supervision and a lot of flexibility and agility [2]–[4], resulting in greater production and resource efficiency. On the other hand, this integration exposes SCADA-based IIoT networks to serious security threats and vulnerabilities, posing a significant danger to these networks and the trustworthiness of the systems [5]. The trustworthiness of an IIoT-enabled system ensures that it performs as expected while meeting a variety of security requirements, including trust, security, safety, reliability, resilience, and privacy [6]–[8]. Fig. 1 depicts the fundamental aspects of trustworthiness in an IIoT-enabled network. The basic goal of the IIoT-enabled system is to increase trustworthiness by safeguarding identities, data, and services, and therefore to secure SCADA-based IIoT networks from cybercriminals [8], [9].

Conclusion

     The ability to protect SCADA-based IIoT networks against cyberattacks increases their trustworthiness. The existing security methods along with machine learning algorithms were inefficient and inaccurate for protecting IIoT networks. In this article, we proposed a cyberattacks detection mechanism using enhanced deep and ensemble learning in a SCADA-based IIoT network. The proposed mechanism is reliable and accurate because an ensemble detection model was built using a combination of the PRU and the DT. The proposed method was evaluated across 15 datasets generated from a SCADA-based network, and a considerable increase in terms of classification accuracy was obtained. Compared to state-of-the-art techniques, the obtained outcomes of our method exhibited a good balance between reliability, trustworthiness, classification accuracy, and model complexity, resulting in improved performance.

     In the future, we will employ more powerful deep learning models to further improve trustworthiness by detecting cyberattacks accurately. In addition, we will try to formulate and assess its performance in real-world scenarios. Also, we will work on the selection of optimal features in scenarios when the features are not sufficient.

Abstract

     Providing pay-as-you-go storage and computing services have contributed to the widespread adoption of cloud computing. Using virtualization technology, cloud service providers can execute several instances on a single physical server, maximizing resource utilization. A challenging issue in cloud data centers is that available resources are rarely fully utilized. The server utilization rate is poor and often below 30%. An accurate host workload prediction enhances resource allocation resulting in more efficient resource utilization. Recently, numerous methods based on deep learning for predicting cloud computing workload have been developed. An efficient strategy must predict long-term dependencies on nonstationary host workload data and be quick enough to respond to incoming requests. This study employs a Bidirectional Gated-Recurrent Unit (BiGRU), Discrete Wavelet Transformation (DWT), and an attention mechanism to improve the host load prediction accuracy. DWT is used to decompose input data into sub-bands with different frequencies and to extract patterns from nonlinear and nonstationary data in order to improve prediction accuracy. The extracted features are fed into BiGRu to predict future workload. The attention mechanism is used in order to extract the temporal correlation features. This hybrid model was evaluated with cluster data sets from Google and Alibaba. Experimental results reveal that our method improves prediction accuracy by 3% to 56% compared to a variety of state-of-the-art methods.

Introduction

     Cloud computing indicates the on-demand accessibility of computer system resources, particularly data storage and computing resources, enabling the users to manage without direct intervention [1], [2]. Organizations can rent cloud computing services as an alternative to investing in their own computing infrastructure or data centers. [3]. The various pay-as-you-go cloud services not only enable clients to purchase resources on-demand [4] but also enables the provision of an infinite amount of resource capacity (e.g., CPU, memory, network, and disk) at a reasonable price without investing in infrastructure or incurring additional expenditures for maintenance [5], [6], [7]. The average capacity utilization rate for regular deployments is less than 40%, although businesses require a relatively large number of servers and other resources to ensure the quality of service (QoS) during peak periods [8], [9]. Fig. 1 depicts boxplots of CPU consumption over two working days for 50 Google cluster machines. Each record of this data represents the cumulative consumption in 5 min. As seen in Fig. 1, however, the average CPU utilization rarely exceeds 50%, and in most cases, it is less than 30%.

Conclusion

     This article proposes a host workload prediction method in cloud computing by combining the DWT, BiGRU model and attention mechanism. In addition to learning long-term dependencies in BiGRU, DWT can decompose nonlinear and nonstationary data into predictable subbands in order to predict future host workload in cloud computing. The proposed approach was evaluated using two real-time host load trace datasets, the Google Cluster Database and the Alibaba Cluster. According to the experimental results, basic techniques cannot learn nonlinear data, mainly when random fluctuations occur in the data. However, the model presented in the proposed method shows good compatibility and achieves better results than DWT-LSTM, DWT-BPNN, DWT-SVR, LSTM, BPNN, and SVR in both datasets. Since distributed computing is increasingly oriented toward lightweight virtualization technologies such as containers, we will predict container workloads in the Docker and Kubernetes environments in future research. We run various applications in these environments, and a series of simulations will run to generate tasks and measure the container’s CPU load. Then, the workload will be predicted using machine learning models, and resource provisioning will be performed.

Abstract

In machine learning, two approaches outperform traditional algorithms: ensemble learning and deep learning. The former refers to methods that integrate multiple base models in the same framework to obtain a stronger model that outperforms them. The success of an ensemble method depends on several factors, including how the baseline models are trained and how they are combined. In the literature, there are common approaches to building an ensemble model successfully applied in several domains. On the other hand, deep learning-based models have improved the predictive accuracy of machine learning across a wide range of domains. Despite the diversity of deep learning architectures and their ability to deal with complex problems and the ability to extract features automatically, the main challenge in deep learning is that it requires a lot of expertise and experience to tune the optimal hyper-parameters, which makes it a tedious and time-consuming task. Numerous recent research efforts have been made to approach ensemble learning to deep learning to overcome this challenge. Most of these efforts focus on simple ensemble methods that have some limitations. Hence, this review paper provides comprehensive reviews of the various strategies for ensemble learning, especially in the case of deep learning. Also, it explains in detail the various features or factors that influence the success of ensemble methods. In addition, it presents and accurately categorized several research efforts that used ensemble learning in a wide range of domains.

Introduction

In a world full of diverse and varied data sources. Machine learning has become one of the most important and dominant branches of artificial intelligence methods, which is applied in many fields. There are many different learning algorithms and methods. Each method’s pitfalls and drawbacks are measured in terms of several factors, including performance and scalability. Based on a lot of research in machine learning, two methods dominate learning algorithms; namely deep learning (Deng et al., 2014) and ensemble learning (Polikar, 2012, Sagi and Rokach, 2018, Rokach, 2019). The deep learning techniques can scale and handle complex problems and offer an automatic feature extraction from unstructured data(Kamilaris and Prenafeta-Boldú, 2018). Also, deep learning methods contain several types of network architectures for different tasks, such as feeding forward neural networks (Bebis and Georgiopoulos, 1994), convolutional neural networks (Collobert and Weston, 2008), recurrent neural networks (Yu et al., 2019). Many others (Ain et al., 2017). However, the training process of deep learning models requires a massive effort, and tuning the optimal hyper-parameters requires expertise and extensive trial, which is a tedious and time-consuming task. Also, training more complex deep neural network increases the chance of overfitting.

Conclusion

In machine learning, reducing the bias and the variance of models is one of the key factors determining the success of the learning process. In the literature, it has been proven that merging the output of different classification algorithms might decrease the generalization error without increasing the variance of the model. The previous is the key essence of the so-called ensemble learning. Numerous research efforts have preferred ensemble learning over single-model learning in various domains. The main advantage of ensemble learning is combining several individual models to improve prediction performance and obtain a stronger model that outperforms them. In the literature, there are several ensemble techniques to boost classification algorithms. The main difference between any two ensemble methods is training the baseline models and how to combine them. Several research efforts introduced ensemble learning into deep learning models to remedy the problems appearing during the learning process of deep learning models. Usually, the main challenge of deep learning models is that they need a lot of knowledge and experience to tune the optimal hyperparameters aiming at reaching a global minimum error. However, finding the optimal hyperparameters requires an exhausting technique in the search space, which in turn becomes a tedious and time-consuming task. Thus, several research efforts have applied deep ensemble learning in many fields, and most of these efforts are articulated around simple ensemble methods. This paper provided a comprehensive review of the various strategies for ensemble learning, especially in the case of deep learning. The paper also illustrated the recent trends in ensemble learning using quantitative analysis of several research papers. Moreover, the paper offered various factors that influence ensemble methods’ success, including sampling the training data, training the baseline models, and the fusion techniques of the baseline models. Also, the papers discussed the pros and cons of each ensemble method. Additionally, the paper extensively introduced and presented several research efforts that used ensemble learning in a wide range of domains and categorized these efforts into either traditional machine or deep learning models as baseline classifiers. It is worth noting that an ensemble of deep learning models using simple averaging methods is not a smart choice and is very sensitive to biased baseline models. On the other hand, Injecting diversity in ensemble deep learning can become robust to the biased baseline models. The diversity can be achieved by training different baseline deep learning architectures over several data samples. The diversity, however, is limited by the computation cost and the availability of suitable data to be sampled.

Abstract

     A major problem with deep generative models is verifying that the generated distribution resembles the target distribution while the individual generated sample is indistinguishable from the original data. In particular, for application in astrophysics we need to be sure that the generated data matches our prior knowledge and that the generated samples entail all object types with the correct frequency and diversity. We currently lack objective ways to systematically assess these quality aspects, where human inspection reaches its limits, as this requires detailed analysis of a large data volume. In this work, we identify reasonable metrics for the quality of galaxy image generators. To this end, we compare a small set of conditional image generators, trained on galaxy images with classification labels for visual morphology features. Our main contribution is a new set of cluster-based metrics for matching the generated distribution to the target distribution. Furthermore, we use the Wasserstein distance on proxies for galaxy morphology as well as a number of other metrics commonly used for image generators. The newly introduced cluster-based metrics are good proxies for the quality of the generated distribution and are suited for automatized identification of mode collapse. Furthermore, the cluster metrics allow for a qualitative interpretation of the generated distribution. The metrics based on morphological statistics provide a useful tool to probe the physical soundness of generated samples. Finally, we find that kernel inception distance used with an InceptionV3 model pre-trained on ImageNet is a good proxy for the overall quality of galaxy image generators, although it cannot be interpreted that easily.

Introduction

     Upcoming astronomical surveys with telescopes, such as Euclid (Laureijs et al., 2012) and LSST (Abell et al., 2009), will provide a wealth of data with billions of galaxy images. These hold unprecedented insight in highly researched astrophysical and cosmological questions, such as formation and evolution of galaxies, the cosmic distribution of dark matter, as well as the expansion history of the Universe (Laureijs et al., 2011). However, the huge amount of images is far too big to be investigated by astrophysicists on an individual basis. Instead, fast and systematic extraction of galaxy properties from their images is required in order to inform and constrain physical models. There are a number of computational tools that have been developed and are already in use by the astrophysical community (Rodriguez-Gomez et al., 2019, Boquien et al., 2019, Shamir, 2011). Though these allow for a systematic extraction of properties, they require too much computational resources to provide big collections of high quality mock images in a reasonable amount of time. Hence, it is necessary to replace these tools by faster methods, e. g. by using machine learning techniques, such as deep neural networks (Lovell et al., 2019, Ferreira et al., 2020, Walmsley et al., 2022 e. g.).

Conclusions

     In order to find suitable evaluation metrics to assess the quality of galaxy image generators, we investigate a number of evaluation metrics with the potential to measure different quality aspects. These metrics are probed with a small number of conditional generative models, some of which purposely are of worse quality than the others. We identified evaluation metrics that are good proxies for the quality of individual images and the resemblance of the target distribution. Our main results are:

• The newly introduced cluster-based metrics (Section 3.1) are a formidable new tool to assess the distribution of generated data.

• The KID metric on features of the pre-trained InceptionV3 model (Szegedy et al., 2015) provides a useful proxy on the overall quality of galaxy image generators.

• Using an ALReLU activation (Mastromichalakis, 2020) significantly enhances classification accuracy for rare object types. This is required to train conditional generators on the highly imbalanced dataset of galaxy images.

     We introduce a new set of cluster-based metrics (Section 3.1), which are good proxies for the distribution quality. In particular, they allow for a qualitative interpretation regarding the sample diversity and the amount of samples generated for different types. Thus, they provide an intuitive new tool to assess the resemblance of the target distribution in a qualitative way, which so far has been a rather illusive task. In addition, one of the cluster metrics, the cluster error provides a formidable tool to identify mode collapse. Moreover, the cluster metrics have the potential to be transformed into a loss function to train generative models to reproduce the distribution of training data exactly, or any other target distribution, e. g. balanced datasets. Together, the aforementioned metrics provide a good basis to assess the quality of trained image generators, which is not necessarily limited to the context of galaxy images. We are currently working on providing losses based on these cluster metrics, which have the potential to drastically improve training of generative models regarding distribution quality, which is especially valuable for GANs.

Abstract

In recent years, people got more interested in the research of the evolution of human-computer interaction (HCI). Right from the transition of a text-based command line interface to a graphical user interface (GUI) that uses a mouse, many changes have been observed. The ability to manipulate the mouse with hand movements is one of the areas of interest here. At present, due to the cluttered backgrounds and poor lighting, this is a complicated issue. The application demands extremely specific detection and recognition. In this article, a computer vision-based method has been shown for controlling a computer mouse with hand gestures in front of a linked camera. The user may move the mouse and use the left as well as right click functions using this systematic program.

INTRODUCTION

In today’s world, computers have become a day-to-day necessity for humans even for the smallest tasks. To make our work easier, it needs to interact with computer ample number of times a day, to make our work easier. This led to the emergence of a hot topic for research, the Human Computer Interaction. A computer’s mouse is extremely helpful tool that makes it much easier for humans to interact with the machine. But the problem occurs when a person with a physical disability uses the mouse and as a result, while using a mouse can be challenging in situations, it may be impossible for others depending on the severity of the disability. To solve this problem, a system that uses a gesture recognition technology is developed to provide mouse control (GRS). The inconvenience of using a mouse while travelling is another issue that will be addressed while developing GRS, thus a better mechanism must be put in place to manage it. Humans rely heavily on their senses for their communication. Similar to this, a computer’s Graphical User Interface (GUI) heavily relies on the mouse.

By blending both of these approaches in the field of digital signal processing, a more effective interactive system for human-computer interaction is formed[8][9][10]. The quantity of information included in an image signal can be used to control a variety of computer operations. Hand gestures are therefore vital in this subject. In this research, a method  for  controlling  the  mouse  with  hand  gestures  is offered in front of a system camera utilizing computer vision. Python libraries are employed, which are excellent tools for working with images and modifying them as well as automating computer programs like OpenCV and patagium. The user needs only the laptop’s web camera to use this; no colored markers are required. For hand detection, this method does not require any dots or bands on the fingers or wrists. The user intends to scroll the mouse in the direction of the fingers’ movement when three fingers are raised and spaced apart. If the fingers are pressed together, the user can move his hand without moving the mouse if he/she chooses. The user must lift 4 fingers to click left and 5 fingers to click right, respectively. In this article, a thorough study has been done on literature, a detailed description is given in dataset portion, the mathematical formulation of model has been done, along with the methodology and experimentation results has been discussed in this paper. Last but not the least the paper is concluded in an effective manner.

Abstract

Among the most prominent field in the human-computer interface (HCI) is emotion recognition using facial expressions. Posed variations, facial accessories, and non-uniform illuminations are some of the difficulties in the emotion recognition field. Emotion detection with the help of traditional methods has the shortcoming of mutual optimization of feature extraction and classification. Computer vision (CV) technology improves HCI by visualizing the natural world in a digital platform like the human brain. In CV technique, advances in machine learning and artificial intelligence result in further enhancements and changes, which ensures an improved and more stable visualization. This study develops a new Modified Earthworm Optimization with Deep Learning Assisted Emotion Recognition (MEWODL-ER) for HCI applications. The presented MEWODL-ER technique intends to categorize different kinds of emotions that exist in the HCI applications. To do so, the presented MEWODL-ER technique employs the GoogleNet model to extract feature vectors and the hyperparameter tuning process is performed via the MEWO algorithm. The design of automated hyperparameter adjustment using the MEWO algorithm helps in attaining an improved emotion recognition process. Finally, the quantum autoencoder (QAE) model is implemented for the identification and classification of emotions related to the HCI applications. To exhibit the enhanced recognition results of the MEWODL-ER approach, a wide-ranging simulation analysis is performed. The experimental values indicated that the MEWODL-ER technique accomplishes promising performance over other models with maximum accuracy of 98.91%.

Introduction

Since people have become more informed, they need a high level of computer intelligence [1], [2]. Human-computer interaction (HCI) is not limited to original hardware-related communication. Certain smarter communication techniques are appearing gradually in the life of people like a sequence of more intellectual techniques relevant to voice recognition [3], face recognition, and gesture recognition. Intellectual mechanisms can help establish interactions between computers and humans [4], [5]. The advent of more convenient communication techniques is becoming a major advancement trend in the current domain of HCI. The objective of HCI development was naturally to make computers adapt and serve the requirements of individuals [6].

People-centred instead of compelling persons to adapt to computers. Emotions had a main role during the interaction. Detection of facial emotions will be helpful in several tasks like social robots, criminal justice systems, security monitoring, customer satisfaction identification, smart card applications, e-learning, etc [7], [8]. The core blocks in the conventional emotion recognition mechanism were classifying the emotions, detecting faces, and extracting the features [9].

Conclusion

In this study, a novel MEWODL-ER method was introduced for emotion classification in the HCI applications. The presented MEWODL-ER algorithm is intended for the identification of various types of emotions that exist in the HCI applications. The proposed model follows a three stage process namely GoogleNet feature extraction, MEWO based hyperparameter tuning, and QAE classification. The design of automated hyperparameter adjustment using the MEWO algorithm helps in attaining an improved emotion recognition process. To show the enhanced recognition results of the MEWODL-ER approach, a wide-ranging simulation analysis is done. The experimental values indicated that the MEWODL-ER technique accomplishes promising performance over other models with higher accuracy of 98.91%. In the future, the efficiency of the MEWODL-ER algorithm will be boosted by hybrid DL classifiers.

Abstract

1. Introduction

2. Related works

3. Preliminaries

4. Our approach

5. Performance evaluation

6. Discussion

7. Final remarks

Acknowledgments

References

Abstract

     In this paper, we address the issue of disaster damage assessments using deep learning (DL) techniques. Specifically, we propose integrating DL techniques into the Internet of Things Search Engine (IoTSE) system to carry out disaster damage assessment. Our approach is to design two scenarios, Single and Complex Event Settings, to complete performance validation using four Convolutional Neural Network (CNN) models. These two scenarios are designed with three possible network services. Our experimental results confirm that all four CNN models can learn each label during the single event setting well. Whereas, with complex event settings, the CNN models have learning difficulty because multiple events have closely related labels.

Introduction

     As technology advances, so does our ability to provide data and deep learning (DL)-driven analytics and predictions. ML techniques, especially DL, have received growing attention and applied to numerous areas, including image and video classification, natural language processing, robotics, networking, mobile computing and cybersecurity, among others [1–9]. DL is prominent in our daily lives; at home with our digital assistants (Apple’s Siri, Amazon’s Alexa, Google Home), within our businesses that forecast finance models, among social interactions affording collaboration, and around emerging in autonomous vehicles and smart homes with a variety of smart devices (NEST thermostats, Ring cameras) [3,10]. DL techniques work to secure the network with anomaly and intrusion detection systems using supervised and unsupervised techniques, provide the capability of identifying security breaches that can occur on a specific computer or on the network [11–15], and balance privacy, utility, safety and reliability [16,17].

Discussion

     Potential future research directions for DL-based IoTSE for damage assessment concern performance and security.

     Performance issue: After running through the performance evaluation, we have determined that all of these models are less likely to recognise the same label in a different but same-topic event. There are two potential ways to improve this. One is utilising DL models that will be able to understand the same or similar context from different datasets. The other is designing a new data pre-processing scheme, which focuses on the similar topic dataset, making the same label data easier for the model to know whether they are the same thing. With respect to the network performance issue, since the whole IoTSE system may be operated over a constrained network environment, the more queries sent to IoTSE, more overhead occurs. To deal with a large number of queries (the cost for responding some queries could be high), we can not only develop the multi-class scheduling algorithms to handle queries with different priorities but also enhance the design of IoTSE architecture to reduce the time taken for processing queries [40 41]. In addition to use the traditional scheduling and optimisation techniques to efficiently manage resources in IoTSE, the new data-oriented network architectures such as Named Data Network (NDN) [64] could be considered.

Abstract

1. Introduction

2. Related work

3. Phases of filtered deep learning model for intrusion detection

4. Result and discussion

5. Conclusion

Declaration of Competing Interest

References

Abstract

     Distributed Sensor Networks play a vital role in the day-to-day world of computing applications, from the cloud to the Internet of Things (IoT). These computing applications devices are normally attached with the microcontrollers like Sensors, actuators, and Adriano network connectivity. Defensive network with an Intrusion Detection System thus serves as the need of modern networks. Despite decades of inevitable development, the Intrusion Detection System is still a challenging research area as the existing Intrusion Detection System operates using signature-based techniques rather than anomaly detection. The existing Intrusion Detection System are thus facing challenges for improvement in Intrusion Detection, Handling heterogeneous data sources is hard for discovering zero-day attacks in IoT networks. This paper presents Filtered Deep Learning Model for Intrusion Detection with a Data Communication approach. The proposed model is composed of five phases: Initialization of Sensor Networks, Cluster Formation in addition to Cluster Head Selection, Connectivity, Attack Detection, and Data Broker. The proposed Model for Intrusion Detection was found to outperform the existing Deep Learning Neural Network and Artificial Neural Network. Experimental results showed a better result of 96.12% accuracy than the dominant algorithms.

Introduction

     In current years, cloud computing (CC) is getting increasingly well-liked in educational along with commercial places [1], in addition, its function is also largely accepted for data storage along with retrieval on the cloud environment [2]. For the basis of increasing the number of data on the Cloud servers(CS), the cloud storage consent to store data in numerous inaccessible sites typically hold by “top” business and running their owner solutions, e.g., Google Drive, Dropbox, Amazon Simple Cloud Storages Service (S3) [3] that means the data is amassed distributive. Even if numerous attractive advances to data gathering were developed in the past decade, it still stays as a research focus in full sway with several significant challenges. Certainly, the incessant decrease in sensor size along with cost, the diversity of sensors present on the marketplace together with the incredible progress on Sensor Node communication technologies have expanded potentially the effect of IoT [4].

     IoT comprises SN that is able to compute, sense together with communicate the data by particular communication [5]. For sensing, the SN are utilized, in addition, for computation, numerous algorithms are utilized, at last, for the communication, disparate communication protocols are utilized, for instance, Building Automations and Control Network (BACnet), Distributed Network Protocols version 3 (DNP3) [6], Constrained Application Protocol (CoAP) [7], Message Queue Telemetry Transports (MQTT) [8], together with XMPP (Extensible Messaging and Presences Protocol), AQMP, et cetera, [9]. By utilizing these protocols, the data is amassed.

Conclusion

     DC and effective cloud storage is a challenging one. Many existing works are putting their effort to give a better system. This paper presented a DC and effective distributed cloud storage system centered on the IDS using the FDLNN algorithm. For IDS, the proposed system uses the FDLNN algorithm, and for the data broker, uses the LFEHO algorithm, and for DC, the MQTT and AQMP connectivity are used. The MQTT is best for lightweight devices, and the AQMP is best for the heavyweight devices. For the IDS analysis, the TON_IoT datasets is taken and the performance of the FDLNN is compared with the existing DLNN and ANN algorithms based on the precision, recall, F-Measure, and accuracy. Then, the performance of the data broker is analyzed, for which the proposed LFEHO is compared with the existing IANFIS and ANFIS algorithm based on average turnaround time, throughput, process time, response time as well as AWT. The IDS performance is analyzed based on SNs count and the data broker is analyzed grounded on the number of data. For 500 nodes, the FDLNN has 96.12% accuracy. In data broker analysis for all data count, the proposed LFEHO attained better performance. Hence, the proposed system is found to attain excellent performance on considering the existing algorithms.