Minhao Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Minhao Li has a total of 10 co-authors. Here are the top co-authors by number of shared publications:
- Le Wang
Guangzhou University
3 shared publications Binxing Fang
1 shared publication- Minxian Liang
Cyberspace Institute of Advanced Technology, Guangzhou University, Guangzhou 510555, China
1 shared publication Rongxin Hu
1 shared publication- Rongxin Hu
Cyberspace Institute of Advanced Technology, Guangzhou University, Guangzhou 510700, China
1 shared publication Tang Zhou
1 shared publication
Latest Publications
- Open Access
- 30 June 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (13) , 6538
HARMONI: A Two-Stage Hybrid Learning Framework with Dynamic Metric Learning for Interpretable NIDS
The increasing sophistication of cyber attacks has created a growing demand for effective Network Intrusion Detection Systems (NIDSs). Although deep learning has improved NIDS performance, existing models often lack adaptive mechanisms for spatiotemporal feature fusion and struggle with complex traffic distributions characterized by severe intra-class heterogeneity and inter-class overlap. Meanwhile, current interpretability methods mainly rely on feature importance analysis and provide limited insight into the model’s decision process. To address these challenges, we propose HARMONI, a two-stage hybrid learning framework that enhances both detection accuracy and model interpretability. In the first stage, a dual-branch Convolutional Neural Network and Gated Recurrent Unit (CNN-GRU) architecture extracts spatiotemporal features, which are dynamically fused through a lightweight adaptive gating network. The representation learning process is jointly optimized using a Dynamic Class-Center Loss to enforce intra-class compactness and inter-class separability in the latent space. In the second stage, the learned deep representations are concatenated with raw traffic features and fed into an ensemble classifier. This residual-style design mitigates information loss during deep encoding while leveraging the non-linear modeling capability of ensemble learning. We further develop a multi-level interpretability framework based on SHapley Additive exPlanations (SHAP) that analyzes global feature importance, individual feature contributions, and feature interactions to provide quantitative insights into the model’s decision mechanisms. Experiments on four benchmark datasets show that HARMONI consistently outperforms state-of-the-art baselines, achieving 80.19% and 78.24% accuracy on NSL-KDD and UNSW-NB15 respectively, surpassing representative deep learning and ensemble methods.
- 30 June 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (13) , 6538
Open AccessHARMONI: A Two-Stage Hybrid Learning Framework with Dynamic Metric Learning for Interpretable NIDS
The increasing sophistication of cyber attacks has created a growing demand for effective Network Intrusion Detection Systems (NIDSs). Although deep learning has improved NIDS performance, existing models often lack adaptive mechanisms for spatiotemporal feature fusion and struggle with complex traffic distributions characterized by severe intra-class heterogeneity and inter-class overlap. Meanwhile, current interpretability methods mainly rely on feature importance analysis and provide limited insight into the model’s decision process. To address these challenges, we propose HARMONI, a two-stage hybrid learning framework that enhances both detection accuracy and model interpretability. In the first stage, a dual-branch Convolutional Neural Network and Gated Recurrent Unit (CNN-GRU) architecture extracts spatiotemporal features, which are dynamically fused through a lightweight adaptive gating network. The representation learning process is jointly optimized using a Dynamic Class-Center Loss to enforce intra-class compactness and inter-class separability in the latent space. In the second stage, the learned deep representations are concatenated with raw traffic features and fed into an ensemble classifier. This residual-style design mitigates information loss during deep encoding while leveraging the non-linear modeling capability of ensemble learning. We further develop a multi-level interpretability framework based on SHapley Additive exPlanations (SHAP) that analyzes global feature importance, individual feature contributions, and feature interactions to provide quantitative insights into the model’s decision mechanisms. Experiments on four benchmark datasets show that HARMONI consistently outperforms state-of-the-art baselines, achieving 80.19% and 78.24% accuracy on NSL-KDD and UNSW-NB15 respectively, surpassing representative deep learning and ensemble methods.
- 31 December 2025
- Published by Elsevier in Journal of Systems Architecture
- Vol. 173
Unmixing gradients: Uncovering persistent leakage in Federated Learning via Independent Component Analysis
- 31 December 2025
- Published by Elsevier in Journal of Systems Architecture
- Vol. 173
Unmixing gradients: Uncovering persistent leakage in Federated Learning via Independent Component Analysis
- Open Access
- 8 November 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (22) , 11897
MiS-PoW: Mirror-Selected Non-Interactive Proof of Ownership for Cloud Storage
Cloud storage uses proofs of ownership to avoid redundant uploads while keeping file contents secret. Many existing schemes need extra round trips, or rely on predictable sampling. These choices reduce security when an adversary knows part of the file. We present MiS-PoW, a zero knowledge and non-interactive proof of ownership. The protocol derives a synchronized challenge seed from the existing HTTPS/TLS session. The seed binds a discretized time window and the file identifier. Both parties compute the same challenges locally, and the protocol adds no new messages. MiS-PoW samples blocks with a stratified policy without duplicates. The policy enforces coverage across partitions and reduces the advantage of contiguous knowledge and near duplicate files. The proof layer uses STARKs with simple AIR constraints. The constraints check that indices come from the seed, lie in range, are unique, and meet per partition counts. We analyze security and show seed unpredictability, resistance to replay, and bounds under partial knowledge with limited grinding. A prototype shows that verification time does not grow with file size, and proof and bandwidth costs remain modest. MiS-PoW is deployable, privacy preserving, and scalable for cloud storage.
- 8 November 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (22) , 11897
Open AccessMiS-PoW: Mirror-Selected Non-Interactive Proof of Ownership for Cloud Storage
Cloud storage uses proofs of ownership to avoid redundant uploads while keeping file contents secret. Many existing schemes need extra round trips, or rely on predictable sampling. These choices reduce security when an adversary knows part of the file. We present MiS-PoW, a zero knowledge and non-interactive proof of ownership. The protocol derives a synchronized challenge seed from the existing HTTPS/TLS session. The seed binds a discretized time window and the file identifier. Both parties compute the same challenges locally, and the protocol adds no new messages. MiS-PoW samples blocks with a stratified policy without duplicates. The policy enforces coverage across partitions and reduces the advantage of contiguous knowledge and near duplicate files. The proof layer uses STARKs with simple AIR constraints. The constraints check that indices come from the seed, lie in range, are unique, and meet per partition counts. We analyze security and show seed unpredictability, resistance to replay, and bounds under partial knowledge with limited grinding. A prototype shows that verification time does not grow with file size, and proof and bandwidth costs remain modest. MiS-PoW is deployable, privacy preserving, and scalable for cloud storage.