https://www.jdcs.ir/ دوفصلنامه محاسبات و سامانه های توزیع شده fa daily 1 Sun, 23 Aug 2026 00:00:00 +0330 Sun, 23 Aug 2026 00:00:00 +0330 https://www.jdcs.ir/article_247248.html Credit risk is one of the most fundamental challenges faced by financial institutions and banks in the loan approval process. Inaccurate assessment of borrowers' repayment capacity can lead to an increase in non-performing loans and impose significant financial losses on the banking system. Therefore, the application of advanced data analysis techniques and machine learning algorithms for accurate credit risk prediction has gained considerable importance.In this study, a credit risk assessment dataset was utilized, and a comprehensive data preprocessing procedure was performed. This process included handling and imputing missing values, identifying and removing outliers, normalizing numerical features, and encoding categorical variables to improve data quality for model training. Subsequently, three widely used machine learning algorithms, namely Extreme Gradient Boosting (XGBoost), Random Forest, and Support Vector Machine (SVM), were trained to predict loan repayment status.To evaluate the performance of the proposed models, several metrics were employed, including Accuracy, Precision, Recall, F1-Score, and the Area Under the Receiver Operating Characteristic Curve (AUC-ROC), providing a comprehensive and multidimensional comparison of their predictive capabilities. The experimental results demonstrated that the XGBoost model outperformed the other models across most evaluation metrics and exhibited superior ability in correctly identifying both high-risk and low-risk customers. Based on these findings, it can be concluded that gradient boosting–based algorithms, particularly XGBoost, represent an efficient and reliable approach for credit risk prediction in financial institutions. https://www.jdcs.ir/article_247251.html ادغام مدل‌های زبانی بزرگ (LLMs) در لبه پزشکی (Medical Edge)، نیازمند ایجاد یک توازن دقیق بین پاسخگویی آنی و حفظ حریم خصوصی داده‌های بیمار است. اگرچه معماری‌های متمرکز نرخ پردازش (Throughput) بالایی را فراهم می‌کنند، یادگیری فدرال (FL) یک جایگزین محافظ حریم خصوصی را با هزینه کاهش کارایی سیستم ارائه می‌دهد. این مقاله یک تحلیل تجربی و سخت‌افزار در حلقه (HIL) را برای کمّی‌سازی این «بهای حریم خصوصی» ارائه می‌کند. ما فرآیند تنظیم دقیق (Fine-tuning) متمرکز در سرور لبه را با یادگیری فدرال با استفاده از مدل Phi-3-mini به همراه انطباق کم‌رتبه (LoRA) برای تشخیص مراحل جراحی مقایسه می‌کنیم. روش‌شناسی ما فراتر از تخمین‌های تحلیلی رفته و با یکپارچه‌سازی پروفایل‌سازی توان مصرفیِ آنی روی پلتفرم‌های NVIDIA Jetson Orin Nano، اثرات گلوگاه‌های حافظه و کاهش خودکار فرکانس ناشی از حرارت (Thermal Throttling) را لحاظ می‌کند. نتایج نشان می‌دهند که «بهای حریم خصوصی» به صورت گلوگاه‌های ملموس در سطح سخت‌افزار آشکار می‌شود؛ در حالی که مدل‌های متمرکز ۴۳% سریع‌تر همگرا می‌شوند، رویکرد مبتنی بر یادگیری فدرال موجب تاخیرهای هماهنگ‌سازی (Synchronization Delays) قابل‌توجه و ۵% هدررفت انرژی بالاتر می‌شود. ما این توازن‌ها را در قالب یک چارچوب تصمیم‌گیری داده-محور فرموله کرده‌ایم که مرزهای کمّی (مانند تاخیر کمتر از ۱۰ میلی‌ثانیه برای کنترل رباتیک) و معادلات لازم را برای بهینه‌سازی تعادل بین کارایی و حریم خصوصی در اختیار معماران سیستم قرار می‌دهد. این مطالعه به عنوان یک راهنمای بنیادین برای طراحی نسل بعدی سیستم‌های سایبر-فیزیکی پزشکی (CPS) حیاتی-سلامتی محسوب می‌شود. https://www.jdcs.ir/article_247254.html ross-project software defect prediction (CSDP) is one of the difficult tasks in software engineering, particularly when there is limited labeled data for the target project. However, there exist two critical challenges for increasing the model accuracy on this task that are distribution shift among projects and class imbalance. In this work, we propose a novel hybrid model, DDA–SDCCA, based on transfer learning. The model attempts to capture shared feature representations between the source and target domains using Dynamic Distribution Alignment (DDA) andDeep Canonical Correlation Analysis (DCCA). After feature extraction, the SMOTE method is used to balance the data before making final decision by using an ensemble classifier with majority voting. We validated the proposed model on seven projects in NASA datasets and achieved 0.83 AUC and 0.53 F1 with an average, which presents a considerable advantage over baseline approaches. These results demonstrate the effectiveness of our model in improving cross-project defect prediction performance.در این پژوهش، ما یک مدل ترکیبی جدید با نام DDA–SDCCA بر پایه یادگیری انتقالی (transfer learning) پیشنهاد می‌کنیم. این مدل تلاش می‌کند نمایش‌های ویژگی مشترک بین دامنه مبدأ و دامنه هدف را با استفاده از هم‌ترازی پویای توزیع (Dynamic Distribution Alignment - DDA) و تحلیل همبستگی متعارف عمیق (Deep Canonical Correlation Analysis - DCCA) استخراج کند.پس از استخراج ویژگی‌ها، از روش SMOTE برای متعادل‌سازی داده‌ها استفاده می‌شود، و در نهایت تصمیم‌گیری نهایی با استفاده از یک طبقه‌بند ensemble و رأی‌گیری اکثریت (majority voting) انجام می‌گیرد.ما مدل پیشنهادی را روی هفت پروژه از مجموعه‌داده‌های NASA ارزیابی کردیم و به‌طور میانگین به مقدار 0.83 AUC و 0.53 F1 دست یافتیم، که نشان‌دهنده برتری قابل توجه نسبت به روش‌های پایه است.این نتایج اثربخشی مدل ما را در بهبود عملکرد پیش‌بینی نقص نرم‌افزار در سناریوی بین‌پروژه‌ای نشان می‌دهد. https://www.jdcs.ir/article_247250.html اینترنت اشیای صنعتی (IIoT) یکی از ارکان اصلی صنعت 4.0 به شمار می‌رود، اما اتکای آن به معماری‌های متمرکز، آسیب‌پذیری‌های مهمی از جمله نقاط تکین خرابی، دستکاری داده‌ها و نقض حریم خصوصی را به همراه دارد. با توجه به اینکه پیش‌بینی می‌شود تعداد دستگاه‌های اینترنت اشیا در سراسر جهان تا سال 2030 از 40 میلیارد دستگاه فراتر رود، مدل‌های امنیتی سنتی توانایی مقیاس‌پذیری لازم را نخواهند داشت و نیاز به چارچوب‌های غیرمتمرکز بیش از پیش احساس می‌شود. فناوری بلاکچین به‌عنوان یک راهکار تحول‌آفرین مطرح شده است که از طریق قراردادهای هوشمند، تغییرناپذیری، شفافیت و اعتماد توزیع‌شده را فراهم می‌کند.این مقاله یک مرور جامع از پیشرفت‌های اخیر (2022 تا 2025) در زمینه ادغام بلاکچین با زیرساخت‌های اینترنت اشیای صنعتی ارائه می‌دهد. ما به‌صورت نظام‌مند بررسی می‌کنیم که چگونه سازوکارهای مبتنی بر بلاکچین چالش‌های امنیتی چندلایه را برطرف می‌کنند و بر احراز هویت امن دستگاه‌ها، ثبت تراکنش‌های مقاوم در برابر دستکاری و کنترل دسترسی غیرمتمرکز تمرکز دارند. یکی از دستاوردهای این مطالعه، ارزیابی الگوریتم‌های اجماع است که برای محیط‌های صنعتی دارای محدودیت منابع بهینه‌سازی شده‌اند تا مصرف بالای انرژی را کاهش دهند.این مقاله راهکارهای نوظهور، از جمله طرح‌های رمزنگاری سبک‌وزن و معماری‌های سلسله‌مراتبی بلاکچین را دسته‌بندی می‌کند که امکان همکاری امن چندجانبه را بدون نیاز به واسطه‌های متمرکز فراهم می‌سازند. با تلفیق روندهای کنونی پژوهش و شناسایی چالش‌های باز، این مطالعه نقشه راهی برای ایجاد شبکه‌های صنعتی نسل آینده ارائه می‌دهد که مقاوم، خودمختار و مقیاس‌پذیر هستند. https://www.jdcs.ir/article_247249.html Smart contract programming languages represent a critical frontier in blockchain technology, demanding rigorous analysis of language design, security features, and runtime semantics. This survey presents a systematic examination of programming languages used for smart contract development across major blockchain platforms, with emphasis on formal verification capabilities, type system properties, vulnerability resistance, and ecosystem maturity. We analyze Solidity and Vyper (Ethereum Virtual Machine), Rust (Solana, Polkadot, NEAR), Move (Aptos, Sui), Cairo (StarkNet), Plutus and Haskell (Cardano), Michelson (Tezos), and Clarity (Stacks), evaluating each language against dimensions including static analysis properties, gas/weight models, resource management, concurrency semantics, and common vulnerability patterns. When viewed through the lens of High-Performance Computing (HPC), smart contract execution introduces unique challenges related to latency, transaction ordering, and maintaining global state consistency across a decentralized network. Therefore, the design choices within these languages—specifically concerning memory management, parallelism handling, and deterministic execution—mirror core concerns in distributed systems engineering. We analyze languages across major platforms, emphasizing their formal verification capabilities, type system properties, and runtime performance metrics relevant to distributed execution efficiency. https://www.jdcs.ir/article_247252.html Diabetes is a condition characterized by elevated blood glucose levels, which should not be overlooked as it can lead to severe complications and organ damage if left untreated. Effective management of diabetes relies on early prediction. Traditional diagnostic methods can be both financially and temporally burdensome for patients and may also be prone to inaccuracies. To address these challenges, modern approaches are proposed. In this project, we aim to achieve early prediction of diabetes in humans by employing various machine learning techniques. Two classification models, XGBoost and LightGBM, were developed based on the Pima Indians Diabetes Database, with hyperparameter tuning to optimize their performance. Both models were rigorously evaluated using various metrics, including Accuracy, Precision, Recall, F1-score, and False Negative Rate. Ultimately, this study achieved an accuracy of 83% with the LightGBM algorithm. After comparing the models, we observed that both demonstrated considerable accuracy in distinguishing diabetic patients, but the LightGBM model exhibited superior performance compared to XGBoost. Based on these findings, the LightGBM model was selected for the development of a web application prototype using the Django framework, which was subsequently deployed for pilot testing, yielding satisfactory results. Our future endeavors will focus on increasing the model's accuracy and completing the necessary steps for its real-world implementation. https://www.jdcs.ir/article_247868.html Traditional relational database–based search approaches often suffer from performance degradation, high query latency, and limited capability in handling large-scale textual and semantic data. These limitations become more pronounced as the volume and complexity of banking transactions grow.This study proposes the design and implementation of a secure, scalable, and distributed hybrid architecture within the Tejarat Bank transaction ecosystem. The proposed system aims to balance security, efficiency, and intelligent data processing. The architecture begins by separating sensitive and non-sensitive data to reduce exposure risks. Sensitive information is protected through encryption mechanisms and secure storage strategies. To improve retrieval capabilities, a semantic analysis layer based on semantic indexing is introduced, enabling concept-based search instead of traditional keyword-based retrieval.In addition, a distributed architecture is employed in which storage, indexing, and query processing are deployed across multiple independent nodes. This design enhances system availability, fault tolerance, and horizontal scalability while distributing computational workload efficiently. A reactive data integration layer connects structured transaction data with semantic indexes using unique identifiers, ensuring consistent and real-time interaction between components.Experimental evaluation demonstrates that the proposed architecture significantly improves both system performance and retrieval accuracy compared to a conventional Oracle database-based approach. In a pilot implementation involving one million transaction records, the average query response time was reduced from 10,000 milliseconds to 2,000 milliseconds. Furthermore, the distributed design improved system throughput and scalability under high request loads. https://www.jdcs.ir/article_247926.html Zero-day attacks are a critical challenge for Network Intrusion Detection (NIDS) due to the unknown structure of the attack with no predefined signature. Machine learning (ML) and Deep Learning (DL) models have recently emerged as a promising solution for enhancing IDS capabilities by detecting anomalies. On the other hand, Federated Learning has emerged as a groundbreaking paradigm for distributed applications to enhance privacy and bandwidth efficiency. The standard Federated Learning (FL) approach generally shares model parameters, which cannot effectively transfer knowledge about unseen attack behaviors across clients. This paper introduces TransFeL-NIDS, a novel federated zero-day detection method that allows clients to share the latent zero-day signatures extracted from a memory-based Transformer. The latent signatures encode high-level behavioral semantics without disclosing raw packets or sensitive traffic data. Compared to the standard FL approach, this strategy improves cross-client generalization to detect unseen attacks while preserving privacy and bandwidth efficiency. Additionally, the latent signature updates help mitigate the negative effects of imbalanced and non-IID data distributions in federated learning. https://www.jdcs.ir/article_247259.html Multi-omics data analysis is about integrating multiple omics data and create strong and complex models to extract deep relations between an organism's genotype and phenotype. The aim of this study is to propose a novel multi-objective evolutionary massive integration method (EMI) for the classification problem of multi-omics data. In this method, first, we have created massive Support Vector Machine (SVM) based models with random features from each omics data type. Then, using an evolutionary algorithm, a combined model of numerous models is created to satisfy two objectives. The first objective of the proposed method is to improve accuracy of the classification problem and the second objective is to find informative biomarkers related to target phenotype. To validate the proposed method, we have gathered breast cancer multi-omics data from The Cancer Genome Atlas (TCGA). We have used estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2) to define breast cancer subtypes. The goal of this study is to predict these subtypes. The proposed method outperforms state of the art articles in this area of research in terms of accuracy and area under curve. Another contribution of this paper is the introduction of some new biomarkers that related to breast cancer subtypes. https://www.jdcs.ir/article_252881.html Upper ontologies provide shared conceptual foundations for knowledge representation, semantic interoperability, and ontology development. However, established frameworks offer limited support for concepts whose interpretation depends on subjective experience, metaphysical commitments, or transcendent forms of existence. This limitation is particularly important in philosophical, theological, and humanities-oriented domains, where such concepts cannot always be adequately represented through a conventional Physical–Abstract distinction.To address this gap, we analyze four widely adopted Upper Ontologies—DOLCE, SUMO, Cyc, and PROTON—and identify a common limitation in representing subjective, spiritual, and metaphysical concepts within a unified semantic framework. Inspired by the conceptual principles of Mulla Sadra's Transcendent Philosophy (al-Hikmah al-Muta'aliyah), particularly the gradation of existence and the hierarchical organization of reality, we propose PANORA , a multi-layer Upper Ontology that systematically models physical, abstract, subjective, and transcendent domains while preserving their semantic relationships.A preliminary qualitative evaluation based on semi-structured interviews with seven experts suggests that the proposed four-domain framework is conceptually coherent, while also identifying areas that require more precise formalization. The reported percentages represent descriptive patterns within this expert panel and are not presented as evidence of technical or logical validation. A preliminary OWL 2 DL formalization of PANORA and accompanying HTML documentation are provided as supplementary materials. https://www.jdcs.ir/article_253713.html Cognitive architectures are computational frameworks designed to simulate human mental processes such as perception, reasoning, and decision-making. Existing architectures, however, largely prioritize computational efficiency while overlooking the ethical and spiritual dimensions of human thought. This paper introduces RAVA (Revelation-Aware Value Architecture), a cognitive-spiritual architecture for intelligent decision-making grounded in a holistic view of human nature. Its five minds function as coordinated, semi-autonomous units under a shared executive, making RAVA suitable for intelligent-agent design. Drawing on the model of five parallel minds, Theory of Mind, and ethical principles, RAVA seeks to balance cognitive efficiency with spiritual purpose. Unlike classical architectures that treat ethics as an external module, RAVA embeds the spiritual dimension throughout the system. At its core, the Nafs and Taqwa Center (NTC) identifies states of the self (nafs), evaluates intention, and provides ethical guidance, transforming the system from a purely computational decision-maker into a moral-formative companion.RAVA was qualitatively evaluated using Grounded Theory through in-depth interviews with five interdisciplinary experts. Findings indicated the greatest structural similarity with CLARION and LIDA, while RAVA remained distinctive in its teleological orientation, model of the self, spiritually oriented reward function, and dedicated ethical center. Experts rated its formative capacity highest (4.8/5), characterizing it as a step toward a “human-formative architecture.” RAVA is empirically testable and offers an interdisciplinary framework for developing artificial intelligence aligned with civilizational values. https://www.jdcs.ir/article_247257.html قراردادهای هوشمند از ویژگی‌های کلیدی بلاکچین هستند؛ با این حال، رابطی برای تبادل داده‌ها با دنیای خارج همچنان منبع مهمی از آسیب‌پذیری‌ها است. این فراخوانی‌ها، قراردادهای هوشمند را در معرض خطرات مختلفی از جمله حملات بازگشتی، دستکاری داده‌ها و اختلالات سرویس قرار می‌دهند. راه‌حل‌های موجود تنها محافظت جزئی ارائه می‌دهند و نمی‌توانند به طور جامع به این چالش‌ها بپردازند. برای رفع این محدودیت، این مطالعه یک رویکرد ترکیبی بهبود یافته را پیشنهاد می‌کند که با ترکیب ذخیره‌سازی انتخابی درون زنجیره‌ای داده‌های با تغییرپذیری کم و فرکانس بالا، همراه با تجزیه و تحلیل فرکانس در سطح تابع و اعتبارسنجی چند لایه، بر اساس مکانیسم‌های اوراکل موجود ساخته شده است. این رویکرد با هدف کاهش وابستگی به منابع خارجی، بهبود قابلیت اطمینان داده‌ها و افزایش تاب‌آوری قراردادهای هوشمند در برابر شرایط نامطمئن و خرابی‌های احتمالی طراحی شده است. روش پیشنهادی با استفاده از مجموعه داده‌ای شامل بیش از 58000 رکورد فراخوانی خارجی از قراردادهای هوشمند تأیید شده اتریوم ارزیابی شد. نتایج تجربی نشان می‌دهد که با وجود سربار اعتبارسنجی اولیه بالاتر، رویکرد پیشنهادی، قرار گرفتن در معرض خطرات مرتبط با فراخوانی خارجی را تحت شکست احتمالی کاهش می‌دهد.