Byeonghui Jeong

Hello! I am Byeonghui Jeong, a Ph.D. candidate at Dongguk University and a member of the uCloud Computing Lab. I investigate how cloud-native services can remain efficient, reliable, and secure under dynamic and uncertain workloads. Specifically, I focus on elastic systems for containerized microservices, with an emphasis on autoscaling, resource management, and workload uncertainty modeling. Ultimately, my goal is to improve resource utilization and service stability by enabling systems to adapt automatically to runtime demand variations. More recently, I have extended this research toward securing elastic systems against attacks that exploit autoscaling mechanisms.

Email  /  Google Scholar  /  ResearchGate  /  ORCID  /  GitHub

News

• Jul. 2025 Autoscaling survey accepted for publication in Computer Science Review.
• Dec. 2024 ARAScaler accepted for publication in IEEE Transactions on Services Computing.
• Aug. 2024 Received the Best Research Award from Dongguk University.
• Apr. 2024 Received the Best Paper Award at MUE 2024.
• Jan. 2024 Mal3S accepted for publication in IEEE Transactions on Information Forensics and Security.
• Sep. 2023 IReS accepted for publication in IEEE Transactions on Consumer Electronics.
• Jul. 2023 HiPerRM accepted for publication in IEEE Transactions on Cloud Computing.
• Nov. 2022 ProHPA accepted for publication in Neurocomputing.
• Oct. 2021 HyMalD accepted for publication in IEEE Transactions on Industrial Informatics.

Education

Work Experiences

Publications

Ongoing Work

International Journal

PUPA addresses workload uncertainty in proactive autoscaling by using adaptive prediction intervals, improving the stability–efficiency trade-off of containerized microservices under nonstationary cloud-native workloads.

The engagement evaluation framework analyzes eye gaze, blinks, facial expressions, and facial landmarks to identify human factors that affect learner engagement in online learning environments.

EProRM combines proactive autoscaling and computation offloading to improve resource efficiency and service continuity in resource-limited fog computing environments.

This survey provides a structured review of cloud-native autoscaling techniques, covering scaling algorithms, mechanisms, evaluation practices, security threats, mitigation strategies, and open research challenges.

HyMNeT improves AIoT intrusion detection by selecting compact multivariate traffic features through mutual information, maximal information coefficient analysis, and evolutionary threshold optimization.

VeCoEL enhances adversarial malware defense by compressing high-dimensional hybrid-analysis features and applying stacking ensemble learning to preserve sequential semantics while reducing perturbation effects.

DRMS manages digital twin workloads by dynamically selecting reactive or proactive autoscaling mechanisms according to LightTS prediction accuracy, improving resource efficiency and service continuity.

ARAScaler predicts future workloads with ETimeMixer and segments them into burst, nonburst, dynamic, and static states to reduce overload, scaling oscillation, and unnecessary scaling events.

TMaD detects variant and obfuscated malware through a cloud–fog–edge collaborative architecture that combines signature-based detection, static analysis, and dynamic analysis of executable files.

BHAS predicts future resource usage and distinguishes burst from non-burst states to stabilize horizontal autoscaling under irregular workloads.

PreVA forecasts future CPU and memory demand with Multi Bi-GRU and applies rolling updates to resize container resources while maintaining sustainable vertical autoscaling.

FoRES predicts future CPU and memory usage with DeHyFo and applies the SERU scoring function to reduce idle resources and node overload during container scheduling.

IReS monitors container workloads and predicts future computing requirements with DLinear to calculate replica counts for efficient digital twin simulation in cloud environments.

Mal3S extracts static malware features, converts them into multi-view images, and trains Multi-SPP-Net to detect known, variant, and transformed malware in Smart IoT environments.

HiPerRM forecasts future CPU and memory usage with SCINet and RevIN, then coordinates vertical and horizontal pod autoscaling to improve resource efficiency and reduce overload.

EPS-Ran analyzes short-term opcode and API-call sequences with a Bi-GRU and TCN hybrid model to predict future ransomware behavior during behavioral analysis.

ProHPA forecasts CPU and memory usage with attention-based Bi-LSTM and combines vertical and horizontal pod autoscaling to reduce resource over-allocation and prevent overload.

HyMalD combines static opcode analysis and dynamic API-call analysis with Bi-LSTM and SPP-Net to detect and classify obfuscated IoT malware.

2-MaD protects IoT devices in smart-city environments through a two-stage hybrid malware detection scheme that combines static and dynamic analysis with deep learning.

Domestic Journal

Conference

Patents

Academic Activities

Technical Skills

Design and source code from Jon Barron's website