Cluster Computing

Günay Yilmaz, Asuman; Gedikli, Eyüp; Aras, Sefa; Kahraman, Hamdi Tolga

doi: 10.1007/s10586-024-05011-3pmid: N/A

Multi-level thresholding image segmentation (MTIS) becomes a difficult and complex problem as the number of thresholds increases. Therefore, meta-heuristic algorithms (MHS) are generally used to solve MTIS problems. However, many problems are encountered in MHS-based MTIS applications. Optimization studies are carried out using different parameter settings and competing algorithms arbitrarily determined by researchers. A few algorithms were used in the experiments, and the optimum solutions were not investigated sufficiently. Also, the feasible solutions were not investigated, and the stability and computational complexity of the algorithms were not analyzed in depth. To solve these problems, Two-Stage Multilevel Image Segmentation (TSMIS) approach was introduced in this study. In the first phase, competitive algorithms, optimum and feasible solutions were determined for the segmentation problems. In the second phase, algorithms that exhibit competitive convergence performance in finding feasible solutions were investigated and their stability analysis was performed. Thanks to TSMIS, an experimental study procedure was developed that defines minimum search conditions to find optimal and feasible solutions. Standards were defined to ensure fairness among competing algorithms and to identify competitive algorithms. An approach was introduced to analyze the stability of algorithms and reveal their computational complexity. In this study, fifteen images from the USC-SIPI image database and Berkeley Segmentation Dataset, two thresholding functions, ten different number of thresholds, and sixty-eight MHS algorithms were used to test and validate the proposed method. According to the statistical analysis results, 13 of the 68 competing algorithms were found to be competitive. 6 of these competitive algorithms- Path Finder (PF), Yin-Yang-Pair Optimization, Linear Population Size Reduction Adaptive Differential Evolution, Fitness-Distance-Balance Based Manta-Ray Foraging Optimization, Supply–Demand-Based Optimization, and Atom Search Algorithm- were applied for the first time to MTIS problem in this study. The stability and computational complexity of the algorithms were also analyzed for the first time in the study. The proposed approach is a candidate to provide reusable data for the execution of future image segmentation studies and to be a standard approach for meta-heuristic-based MTIS. According to the findings, it was concluded that the minimum value of the maxFEs parameter has changed for different MTIS problems, and the minimum value should be maxFEs = 3000 * number of thresholds.

Sudha, Chinnakka; Bolla, Sreenivasulu

doi: 10.1007/s10586-024-05012-2pmid: N/A

Intrusion forecasting systems are developed to safeguard firm networks from attacks. Intrusion prediction is crucial for effectively recognizing and mitigating security breaches, protecting sensitive information, and ensuring integrity. Current approaches have limitations in intrusion detection (ID), such as distinguishing legitimate and strange events. This is because the existing techniques face challenges in analyzing large data, limitations in identifying the relevant features, improper noise filtering, and lack of adaptability to detect the attack patterns. Hence, to overcome a novel Puffer fish Federated Learning (PFFL) Framework for forecasting and classifying Intrusion is introduced. Initially, the datasets are collected, and noise is filtered by preprocessing. Subsequently, the relevant features are selected from the pre-processed data by Pufferfish optimization. Consequently, puffer fish fitness traces abnormal events, and they are predicted. Finally, the attack is classified. The Intrusion is classified as normal, DoS, probe, U2R, and R2L. The proposed PFFL model functions in the Python system. Eventually, the proposed PFFL framework was assessed using a few performance metrics and attained 99.52% accuracy, 99.51% Precision, 98.44% recall, 98.96% F score, 0.0047 error rate, and 624.86 s for computation time. The results demonstrate the proposed PFFL model's efficacy in ID and classification.

Wang, Debao; Guan, Shaopeng; Sun, Ruikang

doi: 10.1007/s10586-025-05107-4pmid: N/A

High communication costs are a major challenge for Federated Learning (FL). Existing solutions often struggle to balance client availability and model accuracy while reducing communication overhead. In this paper, we propose Weighted Average Federated Learning via Two-Stage Client Sampling (WAFL-TSCS). Firstly, we use a detection signal to filter out available clients, enabling them to receive the global model and train the local model, thereby ensuring the efficiency and smoothness of the model training process. Then, we utilize the Kullback–Leibler divergence principle to evaluate the difference in weight distribution between the client’s local model and the global model, measuring the contribution and selecting clients with higher contributions to upload their model parameters. This approach reduces communication overhead. Finally, during the model parameter aggregation phase, the uploaded model parameters are weighted according to their contribution, achieving a weighted average update of the global model and improving model accuracy. Additionally, we introduce a dynamic adjustment term in the loss function to ensure algorithm convergence. Experimental results on two datasets, MNIST and CIFAR-10, show that WAFL-TSCS reduces communication overhead by 20% and improves model accuracy by over 0.05% compared to algorithms such as FedAvg and FedDM, demonstrating its effectiveness in reducing communication costs.

Odeh, Ammar; Taleb, Anas Abu; Alhajahjeh, Tareq; Navarro, Francisco

doi: 10.1007/s10586-025-05104-7pmid: N/A

The growing complexity of malware, especially polymorphic and obfuscated variants, has exposed significant limitations in traditional detection methods. This study addresses these challenges using memory forensics to detect and classify malware through deep learning algorithms. Memory-based features, including memory pages, threads, open files, user sessions, system calls, and kernel modules, were extracted from memory dumps using the Volatility and Rekall frameworks. Three deep learning models—Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Autoencoders—were applied to analyze the extracted features. The dataset was divided into ten subsets using 10-fold cross-validation to ensure robustness and prevent overfitting. The models’ performance was evaluated using accuracy, precision, recall, and F1-score metrics. The results show that CNNs and RNNs consistently outperformed Autoencoders, with CNNs achieving the highest accuracy of 97.8%. These findings demonstrate the superior effectiveness of CNNs and RNNs in detecting malware using memory-based data. This research establishes deep learning algorithms, particularly CNNs and RNNs, as powerful tools for malware detection in cybersecurity. In conclusion, this study contributes to ongoing efforts to enhance malware detection systems by leveraging memory forensics and deep learning. Future work will explore additional feature extraction techniques and hybrid model architectures to improve detection accuracy further and reduce false positives.

Cortes-Goicoechea, Mikel; Mohandas-Daryanani, Tarun; Muñoz-Tapia, Jose Luis; Bautista-Gomez, Leonardo

doi: 10.1007/s10586-025-05153-ypmid: N/A

Modern public blockchains like Ethereum rely on p2p networks to run distributed and censorship-resistant applications. With its wide adoption, it operates as a highly critical public ledger. On its transition to become more scalable and sustainable, shifting to PoS without sacrificing the security and resilience of PoW, Ethereum offers a range of consensus client implementations to participate in the network. In this paper, we present a methodology to measure the performance of the consensus clients based on the latency to receive messages from the p2p network. The paper includes a study that identifies the incentives and limitations that the network experiences, presenting insights about the latency impact derived from running the different consensus implementations at different locations. Our study highlights the need for a holistic approach to node deployment, where hardware, software, and geographic factors have to be carefully considered. Properly dimensioned hardware is essential to mitigate latency-related performance issues and ensure the reliable operation of beacon nodes, especially in geographically distant locations.

Abdulsatar, Majid; Ahmad, Hussain; Goel, Diksha; Ullah, Faheem

doi: 10.1007/s10586-024-05092-0pmid: N/A

The widespread adoption of microservice architectures has given rise to a new set of software security challenges. These challenges stem from the unique features inherent in microservices. It is important to systematically assess and address these software security issues through effective security risk assessments. However, existing risk assessment approaches, such as expert-based manual assessment, prove inefficient in accurately evaluating the security risks of microservices. Furthermore, the absence of security vulnerability metrics hampers the evaluation of these risks. To address these issues, we propose CyberWise Predictor, a framework designed for predicting and assessing security risks associated with microservice architectures. Our framework employs transformers, which are deep learning-based natural language processing models, to analyze descriptions of vulnerabilities for predicting vulnerability metrics to assess security risks. Our experimental evaluation shows the effectiveness of CyberWise Predictor, achieving an average accuracy of 92% in automatically predicting vulnerability metrics for their risk assessment.

Odeh, John Owoicho; Yang, Xiaolong; Samuel, Oluwarotimi Williams; Dhelim, Sahraoui; Nwakanma, Cosmas Ifeanyi

doi: 10.1007/s10586-025-05114-5pmid: N/A

New trends in network architecture have resulted from the introduction of Internet of Things (IoT)-powered innovations. As an ecosystem of devices, sensors, applications and associated network systems, that connect to collect, monitor and analyze data for industrial operations, the Industrial Internet of Things exemplify how their deployment has led to the interconnection of heterogeneous devices and edge networks. This has raised concerns related to device heterogeneity, security and privacy of network infrastructure, which, if unaddressed, leads to the evasion of user and device privacy. Thus, enhancing user confidentiality, enabling secure communication and control of data flows within the edge networks, are essential. For user and network infrastructure security, privacy preservation strategies, such as anonymization, perturbation, location privacy, and differential privacy have been deployed. The privacy-preservation method demonstrated by Dwork’s differential privacy was found to be limited in protecting the Industrial IoT-enabled physical infrastructure. In this paper, we analyze; privacy protection requirements, the threat model with the edge network infrastructure, limitations, demerits and merits of these techniques. Finally, we investigate privacy protection techniques used within four major components of Industrial IoT enabled, critical edge network infrastructure based application which include; Networks (SDN), Data Analytics (Fog Computing), Smart Grid (Intelligent Sensor) and Applications (Pre-processing). These findings are highlighted, with suggestions of areas for future research.

Wu, Xiaohua; Zhang, Tingbo; Chen, Lei; Wang, Zirui

doi: 10.1007/s10586-024-05039-5pmid: N/A

Most existing blockchain platforms cannot transfer assets and messages with other blockchains, which renders them isolated systems. As a kind of cross-chain mechanism, the notary scheme can introduce a notary to forward transactions between different blockchains to enhance blockchain interoperability. However, the notary scheme lacks privacy-preserving mechanisms, which makes users’ private information vulnerable to malicious attacks. To address this issue, this paper proposes a privacy-preserving notary scheme based on zero-knowledge proofs. We combine the existing BlockMaze scheme with notary schemes to ensure that the sender, recipient, and amount of cross-chain transactions are hidden from outsiders. Furthermore, the notary group will use traceable group signatures to sign on cross-chain transactions. When the received transactions are reported illegally, the regulator group can trace the real identities of malicious notaries through the signatures to protect users’ assets. Finally, we conducted a theoretical analysis of the proposed scheme from the perspectives of correctness, anonymity, and security. The correctness analysis verified the consistency of the keys during the execution of the scheme, the anonymity analysis examined the privacy protection of the sender, receiver, and transaction amounts during the transaction process, and the security analysis evaluated the scheme’s defense capabilities against common attacks. Moreover, experimental results demonstrated that the proposed scheme has significantly improved efficiency compared to existing schemes, particularly regarding computational cost and time consumption when handling large prime numbers.

Huynh, Hoang-Hai; Nguyen, Xuan-Ha; Nguyen, Xuan-Duong; Le, Kim-Hung

doi: 10.1007/s10586-024-05075-1pmid: N/A

Software-defined networking (SDN) offers promising network solutions in a big data environment, but existing network intrusion detection systems (NIDS) are limited in handling the high volume of network traffic data. To address this challenge, we propose an SDN-based architecture designed for efficient big data analysis and enhanced monitoring, seamlessly integrating NIDS. The attack detector of our approach is a hybrid model leveraging the advances of both machine and deep learning paradigms with big data processing technologies; thus, it ensures a high processing rate and accuracy in detecting and classifying cyber attacks. The evaluation results on four popular NIDS datasets show that our system could detect several attacks with an accuracy rate of 99% and maintain a minimal false alarm rate of 0.35%. In addition, in a simulated distributed environment, our proposal could process over 40,000 flows per second using just five worker nodes.

Dhal, Krishna Gopal; Das, Arunita; Ray, Swarnajit

doi: 10.1007/s10586-024-05027-9pmid: N/A

Image segmentation with multi-level thresholding (MLT) has become simple and effective in recent times. However, conventional methods for determining appropriate threshold points are time-consuming. Furthermore, one-dimensional histogram-based thresholding methods ignore the spatial correlation between pixels. So, this paper creates a new two-dimensional (2-D) histogram using morphological reconstruction (MR) to make MLT-based models better at segmentation. The 2-D Rényi entropy is used as an objective function. This study also developed a novel opposition learning-based optimizer (OLBO) for finding the optimal thresholds in a reasonable time. The OLBO is a population-based metaheuristic algorithm that is developed based on different opposition-based learning (OBL) strategies. The proposed OLBO has been applied to find the optimal threshold values by maximizing 2-D histogram-based Rényi entropy. The optimization ability of the OLBO has also been evaluated by comparing it with cutting-edge metaheuristic algorithms over well-known mathematical benchmark functions. The findings over benchmark functions demonstrate that OLBO is a competitive metaheuristic algorithm. The proposed OLBO combined with MR-based 2-D (2DMR-OLBO) histogram has been evaluated numerically and visually over the Berkeley Segmentation Dataset (BSDS300). The experimental findings demonstrate that the 2DMR-OLBO method, as described, yields superior segmentation outcomes in comparison to other 2-D histogram-based metaheuristic approaches that are examined.