International Journal of Computer Science and Technology

International Journal of Computer Science and Technology | Vol. 2, No. 4, April 2021 | pp. 25–32

Research Article

Title: Dynamic Task Offloading and Resource Scheduling in Mobile Edge Computing Using Deep Deterministic Policy Gradient

Names of Authors: Santiago Gomez¹, Valeria Rodriguez², and Mateo Lopez³

Authors’ Affiliations:
¹Department of Systems Engineering, National University of Colombia, Bogotá, Colombia
²Faculty of Engineering, University of the Andes, Bogotá, Colombia
³Department of Computer Science, University of Buenos Aires, Buenos Aires, Argentina

Abstract: Mobile Edge Computing (MEC) reduces computational strain on mobile devices by offloading resource-intensive tasks to local edge nodes. However, highly dynamic environments—characterized by unpredictable user mobility, varying task arrival rates, and fluctuating wireless channel qualities—complicate offloading decisions. Static or heuristic scheduling rules cannot adapt to continuous, multi-dimensional system states, resulting in excessive task drop rates and battery depletion. This paper proposes an autonomous, continuous-action task offloading and resource scheduling framework utilizing the Deep Deterministic Policy Gradient (DDPG) algorithm. The MEC ecosystem maps into a continuous state-action space where the actor network determines fractional offloading ratios and local CPU scaling frequencies, while the critic network evaluates the resulting system reward. The target multi-objective reward function minimizes both processing latency and device energy consumption under strict task completion deadline constraints. The model incorporates continuous changes in transmission bandwidth measured in megabits per second (Mbps) and local CPU workloads measured in cycles per second. Simulation experiments demonstrate that the proposed DDPG-based controller converges efficiently, reducing average task execution delay by 34.8% and device battery consumption by 29.1% relative to conventional genetic and Q-learning approaches. The framework ensures reliable execution for interactive edge applications under non-stationary traffic conditions.

Keywords: Mobile edge computing, Task offloading, Deep reinforcement learning, Continuous action space, Resource scheduling, Energy efficiency

Manuscript Timeline: Received: January 15, 2021; Revised: February 22, 2021; Accepted: March 19, 2021; Published: April 1, 2021

Citation: Gomez, S., Rodriguez, V., & Lopez, M. (2021). Dynamic task offloading and resource scheduling in mobile edge computing using deep deterministic policy gradient. International Journal of Computer Science and Technology, 2(4), 25–32. DOI: 10.46882/2021/IJCST/000016

International Journal of Computer Science and Technology | Vol. 2, No. 3, March 2021 | pp. 17–24

Research Article

Title: Optimizing Deep Neural Network Inference on Heterogeneous Edge Devices Using Adaptive Model Quantization

Names of Authors: Chen Wei¹, Li Na², and Zhang Wei³

Authors’ Affiliations:
¹Department of Computer Science and Technology, Tsinghua University, Beijing, China
²State Key Laboratory of Computer Architecture, Institute of Computing Technology, CAS, Beijing, China
³School of Electronic and Information Engineering, Xi'an Jiaotong University, Xi'an, China

Abstract: Deploying state-of-the-art deep neural networks on edge devices enables real-time, low-latency computer vision and natural language applications. However, modern models possess millions of parameters, causing severe memory footprint and power consumption challenges on resource-constrained embedded hardware. Model compression via uniform quantization reduces precision from 32-bit floating-point (FP32) to 8-bit integers (INT8), but often incurs unacceptable accuracy drops in complex network topologies. This study proposes an adaptive, mixed-precision model quantization framework designed to optimize deep learning inference on heterogeneous edge processors. The system uses a layer-wise sensitivity analysis based on Hessian trace estimation to determine the structural tolerance of individual network layers to precision loss. Layers displaying high sensitivity maintain 8-bit or 16-bit precisions, while noise-tolerant components compress to ultra-low 4-bit or 2-bit formats. A multi-objective optimization engine balances model accuracy, execution latency, and memory consumption across target hardware architectures. Hardware experiments on NVIDIA Jetson Nano and Raspberry Pi 4 platforms show that the mixed-precision framework reduces model memory size by 73.5% and accelerates inference execution speed by 2.4 times compared to baseline FP32 implementations. Crucially, top-1 accuracy degradation on the ImageNet dataset is restricted to less than 0.65%, making it highly effective for real-time edge intelligence.

Keywords: Deep learning inference, Edge computing, Model quantization, Mixed-precision, Hardware acceleration, Computer vision

Manuscript Timeline: Received: December 05, 2020; Revised: January 18, 2021; Accepted: February 15, 2021; Published: March 1, 2021

Citation: Wei, C., Na, L., & Wei, Z. (2021). Optimizing deep neural network inference on heterogeneous edge devices using adaptive model quantization. International Journal of Computer Science and Technology, 2(3), 17–24. DOI: 10.46882/2021/IJCST/000015

International Journal of Computer Science and Technology | Vol. 2, No. 2, February 2021 | pp. 9–16

Research Article

Title: An Automated Microservice Decomposition Approach for Monolithic Applications Using Evolutionary Clustering

Names of Authors: Liam Walker¹, Olivia Davies², and Ethan Taylor³

Authors’ Affiliations:
¹School of Computer Science, University of Manchester, Manchester, United Kingdom
²Department of Software Engineering, University of Bristol, Bristol, United Kingdom
³Department of Computer Science, University of Edinburgh, Edinburgh, United Kingdom

Abstract: Modern software engineering favors microservice architectures over monolithic structures to improve scalability, maintainability, and continuous deployment velocity. However, migrating legacy monolithic applications to microservices is a labor-intensive, error-prone manual task that requires deep domain knowledge. Automatic decomposition techniques help, but traditional static code analysis often misinterprets runtime structural dependencies, creating highly coupled services. This paper introduces an automated framework for monolithic application migration that combines static metric extraction with runtime business logic tracing. The methodology maps a legacy system as a multi-layered dependency graph where classes represent nodes, and edges model static method calls, database foreign keys, and shared data structures. A multi-objective evolutionary clustering algorithm based on NSGA-II partitions this graph, maximizing internal service cohesion while minimizing inter-service coupling. Empirical evaluations on three open-source monolithic enterprise applications demonstrate that the framework produces microservice boundaries with an average cohesion improvement of 31.4% and a coupling reduction of 27.2% compared to traditional K-means and hierarchical clustering techniques. The resulting service topologies reduce network communication overhead and respect data transactional boundaries. This approach streamlines legacy software modernization workflows by reducing manual architectural redesign efforts.

Keywords: Microservices migration, Software architecture, Monolithic decomposition, Evolutionary clustering, Multi-objective optimization, Static analysis

Manuscript Timeline: Received: November 15, 2020; Revised: December 20, 2020; Accepted: January 14, 2021; Published: February 1, 2021

Citation: Walker, L., Davies, O., & Taylor, E. (2021). An automated microservice decomposition approach for monolithic applications using evolutionary clustering. International Journal of Computer Science and Technology, 2(2), 9–16. DOI: 10.46882/2021/IJCST/000014

International Journal of Computer Science and Technology | Vol. 2, No. 1, January 2021 | pp. 1–8

Research Article

Title: A Quantum-Resistant Lattice-Based Cryptographic Protocol for Secure Vehicle-to-Everything Communications

Names of Authors: Hans Müller¹, Lukas Schmidt², and Emma Fischer³

Authors’ Affiliations:
¹Department of Mathematics and Computer Science, Technical University of Berlin, Berlin, Germany
²Department of Information Security, Karlsruhe Institute of Technology, Karlsruhe, Germany
³Institute of Computer Science, University of Göttingen, Göttingen, Germany

Abstract: Vehicle-to-Everything (V2X) communication networks enable real-time coordination among autonomous vehicles, roadside infrastructure, and pedestrians to optimize traffic flow and prevent collisions. Securing these safety-critical messages requires high-speed digital signatures to verify sender authenticity and message integrity within milliseconds. However, current V2X security architectures rely heavily on Elliptic Curve Digital Signature Algorithms (ECDSA), which are vulnerable to cryptanalytic attacks by future quantum computers using Shor's algorithm. This paper proposes a quantum-resistant V2X authentication protocol based on the Ring Learning With Errors (R-LWE) hard mathematical problem. The protocol introduces an optimized lattice-based digital signature scheme tailored for the constrained computational environments of On-Board Units (OBUs) in vehicles. To accelerate execution, the Number Theoretic Transform (NTT) is utilized to optimize polynomial multiplication operations. Experimental evaluations conducted on an automotive-grade ARM Cortex-A72 processor reveal that the proposed lattice scheme verifies safety messages in 0.42 ms, well within the 2 ms real-time V2X threshold constraint. The transmission payload increases signature sizes to 1216 bytes, but remains highly manageable over dedicated short-range communications channels. The security analysis proves the scheme achieves existential unforgeability under chosen-message attacks, offering a secure transition path toward post-quantum intelligent transportation systems.

Keywords: Vehicle-to-everything, Post-quantum cryptography, Lattice-based cryptography, Ring learning with errors, Digital signatures, Automotive security

Manuscript Timeline: Received: October 12, 2020; Revised: November 15, 2020; Accepted: December 10, 2020; Published: January 3, 2021

Citation: Müller, H., Schmidt, L., & Fischer, E. (2021). A quantum-resistant lattice-based cryptographic protocol for secure vehicle-to-everything communications. International Journal of Computer Science and Technology, 2(1), 1–8. DOI: 10.46882/2021/IJCST/000013

International Journal of Computer Science and Technology | Vol. 1, No. 12, December 2020 | pp. 89–96

Research Article

Title: Cross-Project Defect Prediction Using Domain Adaptation and Extreme Gradient Boosting

Names of Authors: Min-Ji Kim¹, Seung-Woo Lee², and Ji-Youn Park³

Authors’ Affiliations:
¹Department of Computer Science and Engineering, Seoul National University, Seoul, South Korea
²Department of Software Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South Korea
³Department of Information Technology, Yonsei University, Seoul, South Korea

Abstract: Software defect prediction models assist quality assurance teams by identifying bug-prone source code modules, optimizing testing resource allocation. However, building reliable within-project defect predictors requires extensive historical data, which is unavailable for newly initiated or open-source software projects. Cross-Project Defect Prediction (CPDP) addresses this limitation by using historical data from a source project to train a classifier for a target project. The primary challenge in CPDP is the feature distribution mismatch between the source and target domains, which severely degrades the performance of standard machine learning classifiers. This paper presents a novel CPDP framework that integrates Transfer Component Analysis (TCA) for domain adaptation with an optimized Extreme Gradient Boosting (XGBoost) classifier. The TCA module maps software metrics from heterogeneous source and target projects into a low-dimensional latent space where the distance between data distributions is minimized. The XGBoost model, optimized via random search cross-validation, then executes classification within this aligned feature space. Evaluated on 10 open-source software repositories from the PROMISE dataset, the proposed framework improves the F1-measure by 24.6% and the Area Under the Curve (AUC) by 18.3% compared to non-transfer baseline predictors. The results establish that aligning feature distributions across projects mitigates data scarcity and enhances code auditing efficiency in new software systems.

Keywords: Software defect prediction, Cross-project prediction, Domain adaptation, Transfer component analysis, Extreme gradient boosting, Machine learning

Manuscript Timeline: Received: September 10, 2020; Revised: October 18, 2020; Accepted: November 14, 2020; Published: December 1, 2020

Citation: Kim, M. J., Lee, S. W., & Park, J. Y. (2020). Cross-project defect prediction using domain adaptation and extreme gradient boosting. International Journal of Computer Science and Technology, 1(12), 89–96. DOI: 10.46882/2020/IJCST/000012

International Journal of Computer Science and Technology | Vol. 1, No. 11, November 2020 | pp. 81–88

Research Article

Title: A Distributed Deep Learning Approach for Privacy-Preserving Collaborative Healthcare Diagnostics

Names of Authors: Aris Papadopoulos¹, Chloe Nicolaou², and Dimitris Angelopoulos³

Authors’ Affiliations:
¹Department of Informatics and Telecommunications, National and Kapodistrian University of Athens, Athens, Greece
²Department of Computer Science, University of Cyprus, Nicosia, Cyprus
³School of Electrical and Computer Engineering, National Technical University of Athens, Athens, Greece

Abstract: Deep learning models have demonstrated clinical-grade performance in medical imaging diagnostics, yet training these networks requires massive, highly centralized datasets. In healthcare, centralizing electronic health records or medical scans faces severe regulatory bottlenecks due to strict data privacy legislations like GDPR and HIPAA. This paper proposes a privacy-preserving, distributed deep learning framework based on federated learning (FL) combined with differential privacy (DP) for collaborative disease classification. The proposed system enables multiple independent medical institutions to train a global Convolutional Neural Network (CNN) collaboratively without sharing raw patient data. Each institution computes local model weight updates using its internal dataset, and a central orchestrator aggregates these parameters using the Federated Averaging (FedAvg) algorithm. To thwart membership inference and gradient leakage attacks, a Gaussian noise injection mechanism is integrated into the local gradients based on a rigorous differential privacy budget. Simulations using chest X-ray images for pneumonia detection demonstrate that the proposed framework achieves a diagnostic accuracy of 93.4%, which is within a minor 1.2% margin of the baseline centralized training paradigm. Furthermore, the model successfully maintains data privacy under strong adversarial reconstruction models. The findings demonstrate that combining federated learning with differential privacy provides a secure pathway for large-scale medical AI collaboration without compromising patient confidentiality.

Keywords: Federated learning, Privacy preservation, Differential privacy, Healthcare diagnostics, Collaborative deep learning, Medical imaging

Manuscript Timeline: Received: August 14, 2020; Revised: September 20, 2020; Accepted: October 12, 2020; Published: November 1, 2020

Citation: Papadopoulos, A., Nicolaou, C., & Angelopoulos, D. (2020). A distributed deep learning approach for privacy-preserving collaborative healthcare diagnostics. International Journal of Computer Science and Technology, 1(11), 81–88. DOI: 10.46882/2020/IJCST/000011