Any organization that wants to remain competitive can profit greatly from automating the creation of machine learning systems. To offer models as quickly as possible, many processes and branches are made easier, risk is decreased, and productivity is increased. MLOps, which stands for Machine Learning Operations, is the strategic process that adopts effective techniques for processing, testing, and deploying ML systems in production. It is what makes it all feasible. In this paper, we will study the experimentation of MLOps on the three primary Cloud providers: Microsoft Azure (Azure), Amazon Web Services (AWS), and Google Cloud Platform (GCP). We will then compare the MLOps power of these three platforms.
With the words' current growth in technologies and digitalization, protecting data in massive data-driven systems has become a new challenge to overcome. Therefore, many measures, algorithms, and protocols are used and still under development to minimize risks of Data loss, manipulation, sniffing, or spying… In order to gain more in terms of security, privacy, and integrity levels. As important as this issue is, combined efforts from different disciplines (Network Security, Security Operations Center Analysis (SOC Analysis), Machine Learning (ML), and Deep Learning (DL) algorithms) are necessary to enhance Data protection. In this paper, we will expose some natural threats of Big Data, based on its definition, ethics, and suggestions to avoid these threats, or at least limit their effects. We will also show works related to big data security based on ML and shed light on Artificial Neural Networks (ANN) applications for data classification and threat detection. Then focus on Convolutional Neural Networks (CNN) to optimize parameters. We will try to find solutions for CNN limitations of hyperparameters tuning and rotating image feature extraction. Furthermore, we will discuss state of art results to understand the matter in a deeper sense and expose high-accuracy solutions with improvement path suggestions. Our work is going to come up with some natural threats, compare some existing methods, expose, and discuss results, and then suggest optimal ways to help increase data security, especially through improving CNN limitations.
The rapid expansion of Internet of Things (IoT) devices has resulted in significant progress and developments in various sectors, such as smart healthcare, self-driving vehicles, smart banking, smart home, Industry 4.0, etc. Traditional centralised access control methods are inadequate to deploy in decentralised IoT networks. Although the existing Blockchain-based access control approaches provide a better way of managing access permission for IoT systems, they are ineffective in addressing critical security gaps and preventing unauthorised access while detecting malicious anomalies. To address these two challenges in a single platform, in this paper, we propose a novel approach towards a deep-learning-based authorisation solution that is deployed within the Hyperledger-Fabric (HLF) private Blockchain. Our approach allows smart-contract to define attribute-based access control policies augmented with the Artificial Neural Network (ANN) model, which can effectively identify and isolate malicious anomalies and prevent unauthorised access from malicious devices. We run experiments to evaluate our platform in terms of security and performance, and our results show positive indicators essential for addressing security issues in decentralised IoT networks.
Researchers are concerned about the detection of obfuscated Android malware, and multiple studies have been proposed to address certain obfuscation techniques. However, the comprehensive consideration of all obfuscation techniques remains a critical cybersecurity challenge due to their mutations. To tackle this issue, we developed the DeepMalOb approach, which utilizes memory dumping and deep learning with MLP to detect obfuscated malicious applications. Although the approach has yielded satisfactory results, we acknowledge potential security risks associated with MLPs, such as adversarial attacks, model inversion attacks, overfitting, and model biases, which may impact the accuracy and robustness of the MLP model and render it vulnerable to obfuscated malware. To improve the DeepMalOb approach, we propose the use of formal security risk analysis methods with MLP to detect hidden malware in Android by analyzing the security risks associated with the MLP model and the input features used for training.
Deep learning (DL) is increasingly used in industry, especially in industry 4.0. Thanks to DL, it possible to better prevent breakdowns and manufacturing defects. DL models are becoming more and more complex and efficient, requiring significant compute resources and compute time. The use of Graphic Processing Units (GPUs) makes it possible to speed up processing but at a higher cost. An alternative to them is the use of distributed DL (DDL) which differs from Federated Deep Learning in that it focuses on accelerating calculations and does not address data privacy. DLL requires having several computing nodes. This is where cloud computing comes in. Cloud computing allows resources or virtual machines to be allocated on demand, which reduces costs. However, the allocation of GPU resources has a higher cost than CPU resources, which can be problematic for small businesses. This article proposes to exploit the DDL on CPUs via the on-demand allocation of virtual machines in order to reduce costs. In addition, a solution for deploying the software stack necessary for proper operation is proposed. This is achieved using a containerization which is only composed of the software suites needed to run the DDL to minimize the container transfer size and consequently minimize the container deployment time.
