Yubo Song
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Yubo Song has a total of 122 co-authors. Here are the top co-authors by number of shared publications:
- Bin Xiao
The Hong Kong Polytechnic University
6 shared publications Junbo Wang
Southeast University
5 shared publications- Zhongyuan Qin
School of Cyber Science and Engineering, Southeast University, Nanjing, China
5 shared publications - Liquan Chen
Southeast University
4 shared publications - Shang Gao
Department of Computing, The Hong Kong Polytechnic University, Hong Kong, Hong Kong
4 shared publications - Aiqun Hu
Southeast University
3 shared publications
Latest Publications
- Open Access
- 23 October 2026
- Published by Elsevier in SSRN Electronic Journal
Relation-Constrained Semantic Parsing for Unknown Industrial Control Protocol Analysis
Industrial control protocol messages often contain shared headers while exhibiting function-dependent layouts in downstream fields. Such variations make byte-level similarity insufficient for recovering field structures and semantic roles when protocol specifications are unavailable. This paper formulates unknown industrial protocol analysis as a relation-constrained semantic parsing problem and develops an evidence-driven self-supervised semantic parser for passive traffic analysis. The proposed framework first discovers latent functional message families from observable traffic regularities and uses them as intermediate representations for subsequent parsing. A relation-aware segment graph is then constructed to model candidate field units and integrate structural dependencies within message families, across functional families, and between request-response interactions. The graph-based inference process refines field boundaries by combining local byte evidence with relational constraints. A consistency-aware semantic mapping mechanism further assigns protocol roles to recovered fields based on role-specific evidence and structural dependencies while preserving uncertain fields as unknown. Experiments on Modbus/TCP, IEC 104, MQTT, and S7comm demonstrate that the proposed method achieves a mean V-measure of 0.95 for message-family discovery, a boundary F1-score of 0.91 for field reconstruction, and a semantic macro-F1-score of 0.88 for field role identification.
- 23 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessRelation-Constrained Semantic Parsing for Unknown Industrial Control Protocol Analysis
Industrial control protocol messages often contain shared headers while exhibiting function-dependent layouts in downstream fields. Such variations make byte-level similarity insufficient for recovering field structures and semantic roles when protocol specifications are unavailable. This paper formulates unknown industrial protocol analysis as a relation-constrained semantic parsing problem and develops an evidence-driven self-supervised semantic parser for passive traffic analysis. The proposed framework first discovers latent functional message families from observable traffic regularities and uses them as intermediate representations for subsequent parsing. A relation-aware segment graph is then constructed to model candidate field units and integrate structural dependencies within message families, across functional families, and between request-response interactions. The graph-based inference process refines field boundaries by combining local byte evidence with relational constraints. A consistency-aware semantic mapping mechanism further assigns protocol roles to recovered fields based on role-specific evidence and structural dependencies while preserving uncertain fields as unknown. Experiments on Modbus/TCP, IEC 104, MQTT, and S7comm demonstrate that the proposed method achieves a mean V-measure of 0.95 for message-family discovery, a boundary F1-score of 0.91 for field reconstruction, and a semantic macro-F1-score of 0.88 for field role identification.
- Open Access
- 3 September 2026
- Published by MDPI AG in Electronics
- Vol. 15  (17) , 3977
Study on an Image Candidate Generation Method Based on PCA-Guided Projection
Image retrieval often begins with candidate generation, where a small subset of database images is selected for subsequent reranking or inspection. Standard multi-table p-stable locality-sensitive hashing (LSH) is a classical solution for this stage, but its isotropically sampled projection vectors do not explicitly exploit the principal structure of deep feature distributions. To address this limitation, this paper proposes a PCA-guided candidate generation method using the multi-table bucketization and query architecture of p-stable LSH. Unlike conventional PCA-based feature extraction, PCA is not used here to transform the image features themselves, but to bias the sampling distribution of projection vectors during indexing. Experiments on CALTECH101, CIFAR-10, and Tiny-ImageNet, performed using VGG19, ConvNeXt-Tiny, and ViT-B/16 features, show that the proposed method generally improves on the standard p-stable baseline across different settings. Additional comparisons with HNSW and IVF further position the method relative to modern ANN baselines. The results indicate that PCA-guided projection is an effective lightweight enhancement for systems based on the multi-table p-stable LSH architecture, improving candidate coverage and robustness while remaining compatible with the original online query process.
- 3 September 2026
- Published by MDPI AG in Electronics
- Vol. 15  (17) , 3977
Open AccessStudy on an Image Candidate Generation Method Based on PCA-Guided Projection
Image retrieval often begins with candidate generation, where a small subset of database images is selected for subsequent reranking or inspection. Standard multi-table p-stable locality-sensitive hashing (LSH) is a classical solution for this stage, but its isotropically sampled projection vectors do not explicitly exploit the principal structure of deep feature distributions. To address this limitation, this paper proposes a PCA-guided candidate generation method using the multi-table bucketization and query architecture of p-stable LSH. Unlike conventional PCA-based feature extraction, PCA is not used here to transform the image features themselves, but to bias the sampling distribution of projection vectors during indexing. Experiments on CALTECH101, CIFAR-10, and Tiny-ImageNet, performed using VGG19, ConvNeXt-Tiny, and ViT-B/16 features, show that the proposed method generally improves on the standard p-stable baseline across different settings. Additional comparisons with HNSW and IVF further position the method relative to modern ANN baselines. The results indicate that PCA-guided projection is an effective lightweight enhancement for systems based on the multi-table p-stable LSH architecture, improving candidate coverage and robustness while remaining compatible with the original online query process.
- 2 August 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 333
Structured evidence consolidation for reliable low-resource named entity recognition
- 2 August 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 333
Structured evidence consolidation for reliable low-resource named entity recognition
- 10 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. PP  (5) , 1-13
Traceable, Fair and Privacy-Preserving Decentralized Medical Data Exchange
To harness the commercial potential of medical data, various blockchain-based data-sharing and exchange platforms have been proposed. A key challenge is accurately tracing the provenance and transformations of medical data assets throughout the exchange process. Existing methods cannot facilitate exchanges of publicly hosted datasets while maintaining fairness, as they require revealing keys to the blockchain during interactions. This paper presents MEDET, a novel data exchange scheme that ensures traceability of medical data assets while protecting data privacy and guaranteeing exchange fairness. MEDET leverages zero-knowledge proofs to securely verify transformations within medical datasets and confirm data authenticity. Unlike previous schemes, MEDET supports both simple data exchanges and detailed tracking of data transformations and transaction histories, aiding in the provenance and value assessment of medical records. Additionally, MEDET features a key-secure protocol for fair exchange without disclosing symmetric keys. Compared to existing fair exchange schemes, MEDET uniquely ensures the privacy of publicly hosted data while simultaneously upholding the exchange fairness. The security analysis of MEDET demonstrates its security and privacy properties. The evaluation of MEDET indicates that it outperforms existing schemes in tracking data transformations and facilitating exchanges.
- 10 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. PP  (5) , 1-13
Traceable, Fair and Privacy-Preserving Decentralized Medical Data Exchange
To harness the commercial potential of medical data, various blockchain-based data-sharing and exchange platforms have been proposed. A key challenge is accurately tracing the provenance and transformations of medical data assets throughout the exchange process. Existing methods cannot facilitate exchanges of publicly hosted datasets while maintaining fairness, as they require revealing keys to the blockchain during interactions. This paper presents MEDET, a novel data exchange scheme that ensures traceability of medical data assets while protecting data privacy and guaranteeing exchange fairness. MEDET leverages zero-knowledge proofs to securely verify transformations within medical datasets and confirm data authenticity. Unlike previous schemes, MEDET supports both simple data exchanges and detailed tracking of data transformations and transaction histories, aiding in the provenance and value assessment of medical records. Additionally, MEDET features a key-secure protocol for fair exchange without disclosing symmetric keys. Compared to existing fair exchange schemes, MEDET uniquely ensures the privacy of publicly hosted data while simultaneously upholding the exchange fairness. The security analysis of MEDET demonstrates its security and privacy properties. The evaluation of MEDET indicates that it outperforms existing schemes in tracking data transformations and facilitating exchanges.
- 10 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dependable and Secure Computing
- Vol. 23  (4) , 8360-8375
Fine-Grained IoT Device Fingerprinting Using Active Probing
Existing IoT fingerprinting methods, particularly those based on passive traffic analysis, struggle to distinguish devices with similar firmware and protocol stacks, leading to limited granularity and temporal instability. To overcome these limitations, this paper presents an active probing framework that models each device as a linear time-invariant (LTI) system to capture stable, hardware-level heterogeneity. Maximal-length sequence (MLS) encoded probes serve as broadband excitations, and the echo responses are analyzed in the frequency domain. Leveraging Parseval's theorem, we derive an energy feature that provides a robust and temporally stable device signature, effectively addressing the feature drift problem inherent in traffic based fingerprints. Building on these stable features, an LSH accelerated clustering framework achieves fine-grained, model level identification across 22 IoT devices, successfully distinguishing multiple similar variants from the same vendor. The proposed method achieves 91.3% average accuracy (90.5% mean F1-score) with only 2–4% degradation over 30-day cross-period tests, while reducing clustering time by 26.7%. These results demonstrate a scalable and robust solution for fine-grained IoT device fingerprinting in practical network environments.
- 10 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dependable and Secure Computing
- Vol. 23  (4) , 8360-8375
Fine-Grained IoT Device Fingerprinting Using Active Probing
Existing IoT fingerprinting methods, particularly those based on passive traffic analysis, struggle to distinguish devices with similar firmware and protocol stacks, leading to limited granularity and temporal instability. To overcome these limitations, this paper presents an active probing framework that models each device as a linear time-invariant (LTI) system to capture stable, hardware-level heterogeneity. Maximal-length sequence (MLS) encoded probes serve as broadband excitations, and the echo responses are analyzed in the frequency domain. Leveraging Parseval's theorem, we derive an energy feature that provides a robust and temporally stable device signature, effectively addressing the feature drift problem inherent in traffic based fingerprints. Building on these stable features, an LSH accelerated clustering framework achieves fine-grained, model level identification across 22 IoT devices, successfully distinguishing multiple similar variants from the same vendor. The proposed method achieves 91.3% average accuracy (90.5% mean F1-score) with only 2–4% degradation over 30-day cross-period tests, while reducing clustering time by 26.7%. These results demonstrate a scalable and robust solution for fine-grained IoT device fingerprinting in practical network environments.
- Open Access
- 22 March 2026
- Published by MDPI AG in Electronics
- Vol. 15  (6) , 1324
ConDiffFuzz: Dependency-Aware Consistency Checking for Differential Fuzzing of Industrial Control Protocol Implementations
Consistency checking across independently developed implementations of the same industrial control protocol provides an effective signal for defect discovery because an implementation whose response deviates from the majority under identical inputs is more likely to contain faults or robustness issues. However, existing consistency checking methods remain difficult to apply to complex stateful protocols in practice, since sequence dependencies can cause error propagation, large test suites incur high execution cost across multiple implementations, and inconsistent outputs are costly to triage. This paper proposes ConDiffFuzz, a dependency-aware and dynamically adjusted hierarchical consistency checking method for industrial control protocol implementations. ConDiffFuzz analyzes dependencies among check sequences to optimize execution order and dynamically prunes and regenerates dependent sequences after failures to mitigate inconsistency error propagation. The checking process derives implementation-specific finite state machines and inconsistency records, which further support focused differential fuzzing, parallel execution across multiple implementations, and log-based anomaly triage. Experiments on five Modbus over Modbus/TCP implementations show that ConDiffFuzz achieves a test case acceptance rate of 86.00%, increases average path coverage to 74.46%, improves the average number of triggered anomalies by 12.28%, and reduces the false-positive rate by 20.94% compared with four representative baseline fuzzers (SPIKE, BooFuzz, PeachFuzzer, and Kitty).
- 22 March 2026
- Published by MDPI AG in Electronics
- Vol. 15  (6) , 1324
Open AccessConDiffFuzz: Dependency-Aware Consistency Checking for Differential Fuzzing of Industrial Control Protocol Implementations
Consistency checking across independently developed implementations of the same industrial control protocol provides an effective signal for defect discovery because an implementation whose response deviates from the majority under identical inputs is more likely to contain faults or robustness issues. However, existing consistency checking methods remain difficult to apply to complex stateful protocols in practice, since sequence dependencies can cause error propagation, large test suites incur high execution cost across multiple implementations, and inconsistent outputs are costly to triage. This paper proposes ConDiffFuzz, a dependency-aware and dynamically adjusted hierarchical consistency checking method for industrial control protocol implementations. ConDiffFuzz analyzes dependencies among check sequences to optimize execution order and dynamically prunes and regenerates dependent sequences after failures to mitigate inconsistency error propagation. The checking process derives implementation-specific finite state machines and inconsistency records, which further support focused differential fuzzing, parallel execution across multiple implementations, and log-based anomaly triage. Experiments on five Modbus over Modbus/TCP implementations show that ConDiffFuzz achieves a test case acceptance rate of 86.00%, increases average path coverage to 74.46%, improves the average number of triggered anomalies by 12.28%, and reduces the false-positive rate by 20.94% compared with four representative baseline fuzzers (SPIKE, BooFuzz, PeachFuzzer, and Kitty).
- Open Access
- 13 March 2026
- Published by MDPI AG in Electronics
- Vol. 15  (6) , 1211
Trust Assessment of Distributed Power Grid Terminals via Dual-Domain Graph Neural Networks
As distributed terminals are increasingly integrated into modern power systems with high penetration of renewable energy and decentralized resources, access control mechanisms must support continuous and highly detailed trust assessment. Existing approaches based on machine learning primarily rely on network traffic features from a single source and analyze terminals in isolation, which limits their ability to capture complex device states and correlated attack behaviors. This paper presents a trust assessment framework for distributed power grid terminals that combines multidimensional behavioral modeling with dual domain graph neural networks. Behavioral features are collected from network traffic, runtime environment, and hardware or kernel events and are fused into compact representations through a variational autoencoder to mitigate redundancy and reduce computational overhead. Based on the fused features and observed communication relationships, two graphs are constructed in parallel: a feature domain graph reflecting behavioral similarity and a topological domain graph capturing communication structure between terminals. Graph convolution is performed in both domains to jointly model individual behavioral risk and correlation across terminals. A fusion mechanism based on attention is further introduced to adaptively integrate embeddings specific to each domain, together with a loss function that enforces both shared and complementary representations across domains. Experiments conducted on the CIC EV Charger Attack Dataset 2024 show that the proposed framework achieves a classification accuracy of 96.84%, while maintaining a recall rate above 95% for the low trust category. These results indicate that incorporating multidimensional behavior perception and dual domain relational modeling improves trust assessment performance for distributed power grid terminals under complex attack scenarios.
- 13 March 2026
- Published by MDPI AG in Electronics
- Vol. 15  (6) , 1211
Open AccessTrust Assessment of Distributed Power Grid Terminals via Dual-Domain Graph Neural Networks
As distributed terminals are increasingly integrated into modern power systems with high penetration of renewable energy and decentralized resources, access control mechanisms must support continuous and highly detailed trust assessment. Existing approaches based on machine learning primarily rely on network traffic features from a single source and analyze terminals in isolation, which limits their ability to capture complex device states and correlated attack behaviors. This paper presents a trust assessment framework for distributed power grid terminals that combines multidimensional behavioral modeling with dual domain graph neural networks. Behavioral features are collected from network traffic, runtime environment, and hardware or kernel events and are fused into compact representations through a variational autoencoder to mitigate redundancy and reduce computational overhead. Based on the fused features and observed communication relationships, two graphs are constructed in parallel: a feature domain graph reflecting behavioral similarity and a topological domain graph capturing communication structure between terminals. Graph convolution is performed in both domains to jointly model individual behavioral risk and correlation across terminals. A fusion mechanism based on attention is further introduced to adaptively integrate embeddings specific to each domain, together with a loss function that enforces both shared and complementary representations across domains. Experiments conducted on the CIC EV Charger Attack Dataset 2024 show that the proposed framework achieves a classification accuracy of 96.84%, while maintaining a recall rate above 95% for the low trust category. These results indicate that incorporating multidimensional behavior perception and dual domain relational modeling improves trust assessment performance for distributed power grid terminals under complex attack scenarios.
- Open Access
- 11 March 2026
- Published by MDPI AG in Electronics
- Vol. 15  (6) , 1168
Dynamic Risk Assessment Framework for Concurrent Cyber–Physical Attacks in DER-Integrated Power Grids
Distributed Energy Resource (DER)-integrated power grids are vulnerable to cascading effects under concurrent cyber–physical attacks, where even minor disruptions in system states accumulate and amplify over time, leading to significant system failures. Traditional static risk assessment methods are insufficient for modeling these time-varying, dynamic scenarios, particularly in the context of concurrent attacks. This paper presents a dynamic risk assessment framework leveraging time-synchronized co-simulation, which integrates power system and communication network simulations within a unified time framework. Cyber-attack actions in the communication layer are mapped to corresponding physical disturbances in the distribution network, including voltage, frequency, and power variations. Using the resulting system state evolution trajectories, a Markov Decision Process (MDP)-based state transition tree captures the progression of system risk under concurrent attacks. This framework accounts for cumulative risk across different attack paths and identifies critical nodes and high-risk propagation paths within the network. By incorporating a concurrent event detector into the MDP model, the method quantifies evolving risk dynamics, overcoming the limitations of traditional static methods. Case studies on the IEEE 13-node test feeder and IEEE 14-bus system demonstrate that concurrent attacks result in a security risk metric 2.3 times higher than single-point attacks, validating the effectiveness of the proposed approach in identifying vulnerable nodes whose compromise could lead to cascading failures, supporting the risk-aware prioritization of defensive resources.
- 11 March 2026
- Published by MDPI AG in Electronics
- Vol. 15  (6) , 1168
Open AccessDynamic Risk Assessment Framework for Concurrent Cyber–Physical Attacks in DER-Integrated Power Grids
Distributed Energy Resource (DER)-integrated power grids are vulnerable to cascading effects under concurrent cyber–physical attacks, where even minor disruptions in system states accumulate and amplify over time, leading to significant system failures. Traditional static risk assessment methods are insufficient for modeling these time-varying, dynamic scenarios, particularly in the context of concurrent attacks. This paper presents a dynamic risk assessment framework leveraging time-synchronized co-simulation, which integrates power system and communication network simulations within a unified time framework. Cyber-attack actions in the communication layer are mapped to corresponding physical disturbances in the distribution network, including voltage, frequency, and power variations. Using the resulting system state evolution trajectories, a Markov Decision Process (MDP)-based state transition tree captures the progression of system risk under concurrent attacks. This framework accounts for cumulative risk across different attack paths and identifies critical nodes and high-risk propagation paths within the network. By incorporating a concurrent event detector into the MDP model, the method quantifies evolving risk dynamics, overcoming the limitations of traditional static methods. Case studies on the IEEE 13-node test feeder and IEEE 14-bus system demonstrate that concurrent attacks result in a security risk metric 2.3 times higher than single-point attacks, validating the effectiveness of the proposed approach in identifying vulnerable nodes whose compromise could lead to cascading failures, supporting the risk-aware prioritization of defensive resources.
- 27 October 2025
- Published by Springer Nature in Journal of Network and Systems Management
- Vol. 34  (1) , 1-35
OmniFuzz: A Multi-agent Reinforcement Learning Framework for Protocol-Aware Fuzzing in Power IoT Devices
- 27 October 2025
- Published by Springer Nature in Journal of Network and Systems Management
- Vol. 34  (1) , 1-35
OmniFuzz: A Multi-agent Reinforcement Learning Framework for Protocol-Aware Fuzzing in Power IoT Devices