Van Tong
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Van Tong has a total of 33 co-authors. Here are the top co-authors by number of shared publications:
- Duc Tran
Hanoi University of Science and Technology
10 shared publications - Hai Anh Tran
Hanoi University of Science and Technology
8 shared publications - Nam-Thang Hoang
Faculty of Information Technology, Hanoi University of Civil Engineering, Hanoi, Vietnam
4 shared publications - Ngoc Tuyen Le
Ming Chi University of Technology
4 shared publications - Cong-Son Duong
GRETTIA-COSYS, Universit Gustave Eiffel, Champs-sur-Marne, Marne-la-Valle, France
3 shared publications - Cuong Dao
Hanoi University of Civil Engineering, Hanoi, Vietnam
3 shared publications
Latest Publications
- 12 August 2026
- Published by Springer Nature
- p. 413-426
FusionMalNet: A Hybrid Ensemble Architecture for Windows Malware Detection
- 12 August 2026
- Published by Springer Nature
- p. 413-426
FusionMalNet: A Hybrid Ensemble Architecture for Windows Malware Detection
- Open Access
- 2 October 2025
- Published by Information Security Journal in Journal of Science and Technology on Information security
- p. 62-70
FedPC Cloud CPU Forecasting using Federated Learning and Periodicity-based Clustering
Accurate CPU workload forecasting is vital for efficient resource management and system availability in cloud computing, yet faces challenges in data privacy, security, and the "cold start"problem for new Virtual Machines (VMs). Traditional methods risk privacy and struggle with limited data. We propose FedPC, a novel framework leveraging privacy-preserving Federated Learning (FL) with Periodicity-based Clustering. FedPC enables collaborative training without exposing local data, crucially supporting effective forecasting for new VMs with minimal historical information. It clusters VMs by workload periodicity, training tailored LSTMs within each cluster to handle heterogeneity securely. Evaluated on Azure Public Dataset V1, FedPC surpasses FedAvg in privacy and matches state-of-the-art methods in accuracy. This demonstrates FedPC’s efficacy in securely, adaptively managing resources, thereby enhancing system availability, especially in dynamic cloud environments with frequent VM creation and scarce initial data.
- 2 October 2025
- Published by Information Security Journal in Journal of Science and Technology on Information security
- p. 62-70
Open AccessFedPC Cloud CPU Forecasting using Federated Learning and Periodicity-based Clustering
Accurate CPU workload forecasting is vital for efficient resource management and system availability in cloud computing, yet faces challenges in data privacy, security, and the "cold start"problem for new Virtual Machines (VMs). Traditional methods risk privacy and struggle with limited data. We propose FedPC, a novel framework leveraging privacy-preserving Federated Learning (FL) with Periodicity-based Clustering. FedPC enables collaborative training without exposing local data, crucially supporting effective forecasting for new VMs with minimal historical information. It clusters VMs by workload periodicity, training tailored LSTMs within each cluster to handle heterogeneity securely. Evaluated on Azure Public Dataset V1, FedPC surpasses FedAvg in privacy and matches state-of-the-art methods in accuracy. This demonstrates FedPC’s efficacy in securely, adaptively managing resources, thereby enhancing system availability, especially in dynamic cloud environments with frequent VM creation and scarce initial data.
- Open Access
- 24 August 2025
- Published by Association for Computing Machinery (ACM)
- p. 1815-1817
POSTER: Multimodal Graph Networks for Systematic Generalization in Code Clone Detection
- 24 August 2025
- Published by Association for Computing Machinery (ACM)
- p. 1815-1817
Open AccessPOSTER: Multimodal Graph Networks for Systematic Generalization in Code Clone Detection
- Open Access
- 24 August 2025
- Published by Association for Computing Machinery (ACM)
- p. 21-30
Reducing Communication Overhead in Federated Learning using Deep Reinforcement Learning
- 24 August 2025
- Published by Association for Computing Machinery (ACM)
- p. 21-30
Open AccessReducing Communication Overhead in Federated Learning using Deep Reinforcement Learning
- 8 August 2025
- Published by Springer Nature in International Journal of Information Security
- Vol. 24 (5) , 1-22
PRAU-GIN: GIN-based android malware classification with traffic refinement and node augmentation
- 8 August 2025
- Published by Springer Nature in International Journal of Information Security
- Vol. 24 (5) , 1-22
PRAU-GIN: GIN-based android malware classification with traffic refinement and node augmentation
- 26 April 2025
- Published by Springer Nature
- p. 445-456
A Deep Learning Approach to Early Identification of Remote Access Trojans
- 26 April 2025
- Published by Springer Nature
- p. 445-456
A Deep Learning Approach to Early Identification of Remote Access Trojans
- 14 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Selected Areas in Communications
- Vol. 43 (6) , 2215-2229
Zero Trust: Deep Learning and NLP for HTTP Anomaly Detection in IDS
Web applications have become integral to daily life due to the migration of applications and data to cloud-based platforms, increasing their vulnerability to attacks. This paper addresses the need for robust intrusion detection systems by proposing a system grounded in Zero Trust architecture, which mandates continuous monitoring and multi-layered defenses. The Zero Trust principles ensure ongoing threat assessment and comprehensive protection against various attack vectors. Building on these foundational Zero Trust principles, our study introduces a system designed to not only distinguish normal HTTP requests from well-known attack patterns but also detect emerging types of anomalous attacks. Our system consists of two models that integrate Natural Language Processing approaches, Deep Learning techniques, and Transfer Learning strategies. The first model is employed to detect new anomalous HTTP requests that differ from normal requests. HTTP requests identified as anomalous are transmitted to the second model in charge of classifying specific categories of both well-known and novel attacks. Experiments show that our end-to-end system achieves the average F1-score of 89% on the combination of the CAPEC dataset and the zero-shot CSIC dataset. The proposed system proves also to be able to identify anomalous requests with a minimal latency of 4.8 milliseconds in production settings.
- 14 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal on Selected Areas in Communications
- Vol. 43 (6) , 2215-2229
Zero Trust: Deep Learning and NLP for HTTP Anomaly Detection in IDS
Web applications have become integral to daily life due to the migration of applications and data to cloud-based platforms, increasing their vulnerability to attacks. This paper addresses the need for robust intrusion detection systems by proposing a system grounded in Zero Trust architecture, which mandates continuous monitoring and multi-layered defenses. The Zero Trust principles ensure ongoing threat assessment and comprehensive protection against various attack vectors. Building on these foundational Zero Trust principles, our study introduces a system designed to not only distinguish normal HTTP requests from well-known attack patterns but also detect emerging types of anomalous attacks. Our system consists of two models that integrate Natural Language Processing approaches, Deep Learning techniques, and Transfer Learning strategies. The first model is employed to detect new anomalous HTTP requests that differ from normal requests. HTTP requests identified as anomalous are transmitted to the second model in charge of classifying specific categories of both well-known and novel attacks. Experiments show that our end-to-end system achieves the average F1-score of 89% on the combination of the CAPEC dataset and the zero-shot CSIC dataset. The proposed system proves also to be able to identify anomalous requests with a minimal latency of 4.8 milliseconds in production settings.
- 10 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computational Intelligence
- Vol. 9 (4) , 3147-3161
GAMR: Revolutionizing Multi-Objective Routing in SDN Networks With Dynamic Genetic Algorithms
The growing complexity of modern network systems has increased the need for efficient multi-objective routing (MOR) algorithms. However, existing MOR methods face significant challenges, particularly in terms of computation time, which becomes problematic in networks with short-lived tasks where rapid decision-making is essential. The Non-dominated Sorting Genetic Algorithm II (NSGA-II) offers a promising approach to addressing these challenges. Nevertheless, directly applying NSGA-II in dynamic network environments, where states frequently change, is impractical. This paper presents GAMR, an enhanced non-dominated sorting Genetic Algorithm II-based dynamic multi-objective QoS routing algorithm, which leverages QoS metrics for its multi-objective function. Introducing novel initialization and crossover strategies, our approach efficiently identifies optimal solutions within a brief runtime. Implemented within a Software-defined Network controller for routing, GAMR outperforms existing multi-objective algorithms, exhibiting notable improvements in performance indicators. Specifically, enhancements range from 3.4% to 22.8% on the Hypervolume metric and from 33% to 86% on the Inverted Generational Distance metric. In terms of network metrics, experimental results demonstrate significant reductions in forwarding delay and packet loss rate to 41.25 ms and 3.9%, respectively, even under challenging network configurations with only 2 servers and up to 100 requests.
- 10 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computational Intelligence
- Vol. 9 (4) , 3147-3161
GAMR: Revolutionizing Multi-Objective Routing in SDN Networks With Dynamic Genetic Algorithms
The growing complexity of modern network systems has increased the need for efficient multi-objective routing (MOR) algorithms. However, existing MOR methods face significant challenges, particularly in terms of computation time, which becomes problematic in networks with short-lived tasks where rapid decision-making is essential. The Non-dominated Sorting Genetic Algorithm II (NSGA-II) offers a promising approach to addressing these challenges. Nevertheless, directly applying NSGA-II in dynamic network environments, where states frequently change, is impractical. This paper presents GAMR, an enhanced non-dominated sorting Genetic Algorithm II-based dynamic multi-objective QoS routing algorithm, which leverages QoS metrics for its multi-objective function. Introducing novel initialization and crossover strategies, our approach efficiently identifies optimal solutions within a brief runtime. Implemented within a Software-defined Network controller for routing, GAMR outperforms existing multi-objective algorithms, exhibiting notable improvements in performance indicators. Specifically, enhancements range from 3.4% to 22.8% on the Hypervolume metric and from 33% to 86% on the Inverted Generational Distance metric. In terms of network metrics, experimental results demonstrate significant reductions in forwarding delay and packet loss rate to 41.25 ms and 3.9%, respectively, even under challenging network configurations with only 2 servers and up to 100 requests.
- 27 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 22 (1) , 331-343
Encrypted Traffic Classification Through Deep Domain Adaptation Network With Smooth Characteristic Function
Encrypted network traffic classification has become a critical task with the widespread adoption of protocols such as HTTPS and QUIC. Deep learning-based methods have proven to be effective in identifying traffic patterns, even within encrypted data streams. However, these methods face significant challenges when confronted with new applications that were not part of the original training set. To address this issue, knowledge transfer from existing models is often employed to accommodate novel applications. As the complexity of network traffic increases, particularly at higher protocol layers, the transferability of learned features diminishes due to domain discrepancies. Recent studies have explored Deep Adaptation Networks (DAN) as a solution, which extends deep convolutional neural networks to better adapt to target domains by mitigating these discrepancies. Despite its potential, the computational complexity of discrepancy metrics, such as Maximum Mean Discrepancy, limits DAN’s scalability, especially when applied to large datasets. In this paper, we propose a novel DAN architecture that incorporates Smooth Characteristic Functions (SCFs), specifically SCF-unNorm (Unnormalized SCF) and SCF-pInverse (Pseudo-inverse SCF). These functions are designed to enhance feature transferability in task-specific layers, effectively addressing the limitations posed by domain discrepancies and computational complexity. The proposed mechanism provides a means to efficiently handle situations with limited labeled data or entirely unlabeled data for new applications. The aim is to limit the target error by incorporating a domain discrepancy between the source and target distributions along with the source error. Two statistics classes, SCF-unNorm and SCF-pInverse, are used to minimize this domain discrepancy in traffic classification. The experimental results demonstrate that our proposed mechanism outperforms existing benchmarks in terms of accuracy, enabling real-time traffic classification in network systems. Specifically, we achieve up to 99% accuracy with an execution time of only three milliseconds in the considered scenarios.
- 27 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 22 (1) , 331-343
Encrypted Traffic Classification Through Deep Domain Adaptation Network With Smooth Characteristic Function
Encrypted network traffic classification has become a critical task with the widespread adoption of protocols such as HTTPS and QUIC. Deep learning-based methods have proven to be effective in identifying traffic patterns, even within encrypted data streams. However, these methods face significant challenges when confronted with new applications that were not part of the original training set. To address this issue, knowledge transfer from existing models is often employed to accommodate novel applications. As the complexity of network traffic increases, particularly at higher protocol layers, the transferability of learned features diminishes due to domain discrepancies. Recent studies have explored Deep Adaptation Networks (DAN) as a solution, which extends deep convolutional neural networks to better adapt to target domains by mitigating these discrepancies. Despite its potential, the computational complexity of discrepancy metrics, such as Maximum Mean Discrepancy, limits DAN’s scalability, especially when applied to large datasets. In this paper, we propose a novel DAN architecture that incorporates Smooth Characteristic Functions (SCFs), specifically SCF-unNorm (Unnormalized SCF) and SCF-pInverse (Pseudo-inverse SCF). These functions are designed to enhance feature transferability in task-specific layers, effectively addressing the limitations posed by domain discrepancies and computational complexity. The proposed mechanism provides a means to efficiently handle situations with limited labeled data or entirely unlabeled data for new applications. The aim is to limit the target error by incorporating a domain discrepancy between the source and target distributions along with the source error. Two statistics classes, SCF-unNorm and SCF-pInverse, are used to minimize this domain discrepancy in traffic classification. The experimental results demonstrate that our proposed mechanism outperforms existing benchmarks in terms of accuracy, enabling real-time traffic classification in network systems. Specifically, we achieve up to 99% accuracy with an execution time of only three milliseconds in the considered scenarios.