Jianjin Zhao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Jianjin Zhao has a total of 35 co-authors. Here are the top co-authors by number of shared publications:
- Meng Shen
Beijing Institute of Technology
5 shared publications - Qi Li
School of Cyberspace Security, Beijing University of Posts and Telecommunications, Beijing, China
5 shared publications - Junsong Fu
School of Cyberspace Security, Beijing University of Posts and Telecommunications, Beijing, China
2 shared publications - Zhiwei Cui
Beijing University of Posts and Telecommunications, Beijing, China
2 shared publications - Bharat Bhargava
Department of Computer Science, Purdue University, West Lafayette, IN, USA
1 shared publication - Bowen Sun
School of Cyberspace Security, Beijing University of Posts and Telecommunications, Beijing, China
1 shared publication
Latest Publications
- 24 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 7118-7137
Enhancing Intrusion Detection via Interpretable Inter-Flow Spatio-Temporal Graphs and Intra-Flow Features
The performance of network intrusion detection systems (IDS) critically depends on the completeness of traffic feature representation. Existing methods predominantly focus either on intra-flow features (e.g., packet length sequences) or single-dimensional inter-flow relationships, limiting optimal classification. To address this restriction, we propose IF$^{2}$-STG to enhance intrusion detection via interpretable inter-flow spatio-temporal graphs and intra-flow features. The key innovation of our scheme is the construction of an Inter-flow Spatio-Temporal relationship Graph (ISTG), with two-dimensional edges explicitly quantifying the correlation strength between flows. On this basis, a GNN learns the inter-flow context. In parallel, a CNN extracts fine-grained intra-flow features from raw bytes of individual flows. Finally, the two modal features are fused to acquire a comprehensive representation. This representation can effectively capture both intrinsic behavior patterns of flows and contextual inter-flow relationships, thereby significantly enhancing the detection performance and robustness of model. Experimental results show that the proposed IF$^{2}$-STG achieves over 99% accuracy on three real-world datasets, outperforming existing approaches. The constructed ISTG provides explicit visual interpretation of flow dependencies. More importantly, IF$^{2}$-STG-Inter, using only inter-flow features, maintains 97.78% detection accuracy in cross-dataset tests involving time-disparate datasets, validating superior generalization of inter-flow spatio-temporal features. This advantage renders it particularly promising for practical intrusion detection applications.
- 24 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 7118-7137
Enhancing Intrusion Detection via Interpretable Inter-Flow Spatio-Temporal Graphs and Intra-Flow Features
The performance of network intrusion detection systems (IDS) critically depends on the completeness of traffic feature representation. Existing methods predominantly focus either on intra-flow features (e.g., packet length sequences) or single-dimensional inter-flow relationships, limiting optimal classification. To address this restriction, we propose IF$^{2}$-STG to enhance intrusion detection via interpretable inter-flow spatio-temporal graphs and intra-flow features. The key innovation of our scheme is the construction of an Inter-flow Spatio-Temporal relationship Graph (ISTG), with two-dimensional edges explicitly quantifying the correlation strength between flows. On this basis, a GNN learns the inter-flow context. In parallel, a CNN extracts fine-grained intra-flow features from raw bytes of individual flows. Finally, the two modal features are fused to acquire a comprehensive representation. This representation can effectively capture both intrinsic behavior patterns of flows and contextual inter-flow relationships, thereby significantly enhancing the detection performance and robustness of model. Experimental results show that the proposed IF$^{2}$-STG achieves over 99% accuracy on three real-world datasets, outperforming existing approaches. The constructed ISTG provides explicit visual interpretation of flow dependencies. More importantly, IF$^{2}$-STG-Inter, using only inter-flow features, maintains 97.78% detection accuracy in cross-dataset tests involving time-disparate datasets, validating superior generalization of inter-flow spatio-temporal features. This advantage renders it particularly promising for practical intrusion detection applications.
- 9 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cloud Computing
- Vol. 14 (1) , 224-241
FBFL: Flexible Byzantine-Robust Federated Learning With Privacy-Preserving
Privacy-preserving federated learning allows many clients to collaboratively train a machine learning model by sharing encrypted models. Among them, mask-based schemes have been widely applied due to their efficiency advantages. Unfortunately, as the invisibility of an individual local model, such schemes are vulnerable to poisoning attacks by Byzantine clients. Current work lacks a practical method to detect Byzantine clients within the mask-based scheme. We propose FBFL, a flexible Byzantine-robust federated learning scheme with privacy-preserving. While protecting the privacy of the client, we implement a defense against Byzantine clients without relying on an individual masked local model. Specifically, we design a secure distance computation method based on the Pedersen commitment. The exponential Manhattan distance we design is utilized to compute the malicious score of the client in order to distinguish the benign model from the abnormal model. Since the malicious score is obtained utilizing only the commitment value and not the masked local model, our Byzantine-robust method can be flexibly combined with other mask-based privacy-preserving methods. Security analysis shows that FBFL is Byzantine-robust and can ensure the data security of clients. Experimental evaluation shows that the robustness and efficiency of FBFL are great.
- 9 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cloud Computing
- Vol. 14 (1) , 224-241
FBFL: Flexible Byzantine-Robust Federated Learning With Privacy-Preserving
Privacy-preserving federated learning allows many clients to collaboratively train a machine learning model by sharing encrypted models. Among them, mask-based schemes have been widely applied due to their efficiency advantages. Unfortunately, as the invisibility of an individual local model, such schemes are vulnerable to poisoning attacks by Byzantine clients. Current work lacks a practical method to detect Byzantine clients within the mask-based scheme. We propose FBFL, a flexible Byzantine-robust federated learning scheme with privacy-preserving. While protecting the privacy of the client, we implement a defense against Byzantine clients without relying on an individual masked local model. Specifically, we design a secure distance computation method based on the Pedersen commitment. The exponential Manhattan distance we design is utilized to compute the malicious score of the client in order to distinguish the benign model from the abnormal model. Since the malicious score is obtained utilizing only the commitment value and not the masked local model, our Byzantine-robust method can be flexibly combined with other mask-based privacy-preserving methods. Security analysis shows that FBFL is Byzantine-robust and can ensure the data security of clients. Experimental evaluation shows that the robustness and efficiency of FBFL are great.
- 2 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 245-261
A Unified Framework for Robust Encrypted Malicious Traffic Detection in Adverse Environments via Graph Structure Learning
The widespread adoption of encryption protocols enables attackers to conceal malicious activities within encrypted traffic, rendering traditional detection methods ineffective. Graph Neural Networks (GNNs) have emerged as a promising solution by modeling network objects and their interactions within graph representations to capture the collaborative behavioral patterns of complex threat activities. However, the well-performed premise of GNNs does not always hold in adverse environments, leading to unsatisfactory performance, suffering from three critical issues including (1) incomplete information analysis, where heterogeneous relations among network objects are often overlooked (2) lack of solutions for evasion techniques, as existing methods focus on robust representation learning but fail to correct adversarial distortions, and (3) limited robustness evaluation, relying on synthetic feature perturbations rather than raw traffic manipulations in line with real-world attacks. To address these issues, we propose RETA, a unified framework for robust encrypted malicious traffic detection via graph structure learning. First, RETA unifies heterogeneous subgraphs capturing semantic metapaths and homogeneous subgraphs modeling behavioral similarities among encrypted sessions and takes a tailored Heterogeneous Graph Attention Network (HAN) encoder for neighborhood information aggregation. Then, it employs a unified graph structure learning framework to correct noisy relations induced by evasion techniques through channel attention-based aggregation and Bayesian inference-based estimation. Following an iterative manner, RETA mutually improves relation modeling and detection robustness. Finally, RETA simulates various realistic adverse conditions by modifying raw traffic captures, ensuring comprehensive robustness evaluations against network fluctuations and adversarial attacks. Extensive experiments demonstrate the superior robustness of RETA, significantly improving detection performance in adverse environments. Even under extreme adverse conditions (i.e., 30% packet loss rate and 5 perturbation edges), RETA still shows significant advantages, delivering 8.94% and 4.85% accuracy improvements over the baseline models on average.
- 2 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 13, 245-261
A Unified Framework for Robust Encrypted Malicious Traffic Detection in Adverse Environments via Graph Structure Learning
The widespread adoption of encryption protocols enables attackers to conceal malicious activities within encrypted traffic, rendering traditional detection methods ineffective. Graph Neural Networks (GNNs) have emerged as a promising solution by modeling network objects and their interactions within graph representations to capture the collaborative behavioral patterns of complex threat activities. However, the well-performed premise of GNNs does not always hold in adverse environments, leading to unsatisfactory performance, suffering from three critical issues including (1) incomplete information analysis, where heterogeneous relations among network objects are often overlooked (2) lack of solutions for evasion techniques, as existing methods focus on robust representation learning but fail to correct adversarial distortions, and (3) limited robustness evaluation, relying on synthetic feature perturbations rather than raw traffic manipulations in line with real-world attacks. To address these issues, we propose RETA, a unified framework for robust encrypted malicious traffic detection via graph structure learning. First, RETA unifies heterogeneous subgraphs capturing semantic metapaths and homogeneous subgraphs modeling behavioral similarities among encrypted sessions and takes a tailored Heterogeneous Graph Attention Network (HAN) encoder for neighborhood information aggregation. Then, it employs a unified graph structure learning framework to correct noisy relations induced by evasion techniques through channel attention-based aggregation and Bayesian inference-based estimation. Following an iterative manner, RETA mutually improves relation modeling and detection robustness. Finally, RETA simulates various realistic adverse conditions by modifying raw traffic captures, ensuring comprehensive robustness evaluations against network fluctuations and adversarial attacks. Extensive experiments demonstrate the superior robustness of RETA, significantly improving detection performance in adverse environments. Even under extreme adverse conditions (i.e., 30% packet loss rate and 5 perturbation edges), RETA still shows significant advantages, delivering 8.94% and 4.85% accuracy improvements over the baseline models on average.
- 17 June 2025
- Published by Elsevier in Future Generation Computer Systems
- Vol. 174
One sketch for all super hosts: Large-scale network monitoring with UnivCar
- 17 June 2025
- Published by Elsevier in Future Generation Computer Systems
- Vol. 174
One sketch for all super hosts: Large-scale network monitoring with UnivCar
- 11 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 20 (15566013) , 677-692
ReTrial: Robust Encrypted Malicious Traffic Detection via Discriminative Relation Incorporation and Misleading Relation Correction
Encryption techniques greatly ensure the confidentiality and integrity of network communications. However, they also allow attackers to conceal malicious activities within encrypted traffic, posing severe cybersecurity challenges. Current detection methods primarily rely on statistics and correlation analysis. However, both statistical features and inter-entity relations can be easily obfuscated. Moreover, issues with low-quality data and fixed feature sets limit the generalizability and adaptability to defend against various evasion techniques. Robustifying encrypted malicious traffic detection in adverse conditions is still an open problem. In this paper, we propose ReTrial, a robust encrypted malicious traffic detection system via discriminative relation incorporation and misleading relation correction. The key motivations behind ReTrial are to accurately leverage the rich relations among flows for contextual analysis, and correct misleading ones for robust threat detection. Specifically, we construct a relational multigraph and develop a tailored Graph Attention Network (GAT) to selectively incorporate contextual information. Then we retrieve multi-order neighborhood similarity graphs as observations for adaptive relation correction. Following an iterative scheme, both detector performance and graph topology mutually optimize. To validate the robustness of ReTrial, we simulate various adverse conditions by randomly dropping packets and greedily injecting perturbation edges. The experimental results show that ReTrial is competitive in ideal condition. Under adverse conditions, though the performances of other state-of-the-art methods degrade significantly, ReTrial consistently exhibits superior performance with a maximum reduction of only 5.88% in F1, highlighting its robustness in threat detection.
- 11 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 20 (15566013) , 677-692
ReTrial: Robust Encrypted Malicious Traffic Detection via Discriminative Relation Incorporation and Misleading Relation Correction
Encryption techniques greatly ensure the confidentiality and integrity of network communications. However, they also allow attackers to conceal malicious activities within encrypted traffic, posing severe cybersecurity challenges. Current detection methods primarily rely on statistics and correlation analysis. However, both statistical features and inter-entity relations can be easily obfuscated. Moreover, issues with low-quality data and fixed feature sets limit the generalizability and adaptability to defend against various evasion techniques. Robustifying encrypted malicious traffic detection in adverse conditions is still an open problem. In this paper, we propose ReTrial, a robust encrypted malicious traffic detection system via discriminative relation incorporation and misleading relation correction. The key motivations behind ReTrial are to accurately leverage the rich relations among flows for contextual analysis, and correct misleading ones for robust threat detection. Specifically, we construct a relational multigraph and develop a tailored Graph Attention Network (GAT) to selectively incorporate contextual information. Then we retrieve multi-order neighborhood similarity graphs as observations for adaptive relation correction. Following an iterative scheme, both detector performance and graph topology mutually optimize. To validate the robustness of ReTrial, we simulate various adverse conditions by randomly dropping packets and greedily injecting perturbation edges. The experimental results show that ReTrial is competitive in ideal condition. Under adverse conditions, though the performances of other state-of-the-art methods degrade significantly, ReTrial consistently exhibits superior performance with a maximum reduction of only 5.88% in F1, highlighting its robustness in threat detection.
- 5 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 21 (2) , 2460-2476
MetaRockETC: Adaptive Encrypted Traffic Classification in Complex Network Environments via Time Series Analysis and Meta-Learning
Encrypted Traffic Classification (ETC) is crucial for network security management and Quality of Service (QoS) improvement. There have been many attempts to tackle various ETC tasks, however, which generally suffer from task dependency and limited adaptability, falling short of meeting practical requirements. Under the realistic assumptions of complex network environments, diverse encryption techniques and ever-changing application landscapes coexist. It is highly desirable to learn the generic encrypted traffic representations to investigate the common knowledge across different ETC tasks and rapidly adapt to the dynamic shifts. To fill the gap, we propose MetaRockETC, a generic encrypted traffic classification framework, which extracts protocol-agnostic features to learn the common knowledge and rapidly adapt to novel ETC tasks and evolving network environments. In MetaRockETC, we first model packet length sequences of encrypted sessions as multivariate time series and perform random convolution kernel transformations to summarize discriminatory behavioral patterns across channels. By integrating MetaRockETC into an advanced Model-Agnostic Meta-Learning (MAML) framework, we learn a task-adaptive loss function to facilitate better generalization and transferability across diverse ETC tasks. Extensive experimental results demonstrate the superiority of MetaRockETC in both across-task and few-shot scenarios, highlighting its potential to provide a practical solution for encrypted traffic classification in real-world scenarios.
- 5 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 21 (2) , 2460-2476
MetaRockETC: Adaptive Encrypted Traffic Classification in Complex Network Environments via Time Series Analysis and Meta-Learning
Encrypted Traffic Classification (ETC) is crucial for network security management and Quality of Service (QoS) improvement. There have been many attempts to tackle various ETC tasks, however, which generally suffer from task dependency and limited adaptability, falling short of meeting practical requirements. Under the realistic assumptions of complex network environments, diverse encryption techniques and ever-changing application landscapes coexist. It is highly desirable to learn the generic encrypted traffic representations to investigate the common knowledge across different ETC tasks and rapidly adapt to the dynamic shifts. To fill the gap, we propose MetaRockETC, a generic encrypted traffic classification framework, which extracts protocol-agnostic features to learn the common knowledge and rapidly adapt to novel ETC tasks and evolving network environments. In MetaRockETC, we first model packet length sequences of encrypted sessions as multivariate time series and perform random convolution kernel transformations to summarize discriminatory behavioral patterns across channels. By integrating MetaRockETC into an advanced Model-Agnostic Meta-Learning (MAML) framework, we learn a task-adaptive loss function to facilitate better generalization and transferability across diverse ETC tasks. Extensive experimental results demonstrate the superiority of MetaRockETC in both across-task and few-shot scenarios, highlighting its potential to provide a practical solution for encrypted traffic classification in real-world scenarios.
- 16 August 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11 (3) , 5129-5142
Efficient IoT Device Identification via Network Behavior Analysis Based on Time Series Dictionary
Due to hardware limitations, Internet of Things (IoT) devices without integrated security become easy targets for network attacks. IoT device identification is significant for network security management. Despite many efforts, previous studies either require excessive features raising concerns about efficiency and privacy, or underutilize the data resources to fulfill the potential of simple features. Moreover, the severe data imbalance problem is unaddressed. In this article, we present IoTProfile, an efficient IoT device identification framework via time series dictionary. It only considers simple packet-level attributes and maps them into different time windows. On this basis, it further follows a shuffle&split organization scheme to structure the imbalanced data as multichannel time series. By performing random convolutional kernel transformations in two ways and aggregations, IoTProfile captures discriminative patterns and forms the frequency count of recurring patterns to profile the network behaviors of IoT devices over a period of time. The experimental results show that IoTProfile is superior to the other state-of-the-art methods in terms of both identification effectiveness and time overhead, achieving 99.81% and 97.65% Macro-F1 scores on the University of New South Wales and University of New Brunswick data sets in under 4 min.
- 16 August 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11 (3) , 5129-5142
Efficient IoT Device Identification via Network Behavior Analysis Based on Time Series Dictionary
Due to hardware limitations, Internet of Things (IoT) devices without integrated security become easy targets for network attacks. IoT device identification is significant for network security management. Despite many efforts, previous studies either require excessive features raising concerns about efficiency and privacy, or underutilize the data resources to fulfill the potential of simple features. Moreover, the severe data imbalance problem is unaddressed. In this article, we present IoTProfile, an efficient IoT device identification framework via time series dictionary. It only considers simple packet-level attributes and maps them into different time windows. On this basis, it further follows a shuffle&split organization scheme to structure the imbalanced data as multichannel time series. By performing random convolutional kernel transformations in two ways and aggregations, IoTProfile captures discriminative patterns and forms the frequency count of recurring patterns to profile the network behaviors of IoT devices over a period of time. The experimental results show that IoTProfile is superior to the other state-of-the-art methods in terms of both identification effectiveness and time overhead, achieving 99.81% and 97.65% Macro-F1 scores on the University of New South Wales and University of New Brunswick data sets in under 4 min.