Zhiyi Tian
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Zhiyi Tian has a total of 49 co-authors. Here are the top co-authors by number of shared publications:
- Shui Yu
University of Technology Sydney
27 shared publications - Chenhan Zhang
School of Computer Science, University of Technology Sydney, Sydney, NSW, Australia
21 shared publications - Weiqi Wang
Xi'an Jiaotong University, Xi'an, Shaanxi, China
18 shared publications - An Liu
School of Computer Science and Technology, Soochow University, Suzhou, Jiangsu, China
4 shared publications - Lei Cui
Key Laboratory of Computing Power Network and Information Security, Ministry of Education, Shandong Computer Science Center, Qilu University of Technology (Shandong Academy of Sciences), Jinan, China
4 shared publications - James Jian Qiao Yu
School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen, China
3 shared publications
Latest Publications
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (11) , 19176-19187
VerfSemCom: Deep Learning-Driven Verification for Trustworthy Image Semantic Communication
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (11) , 19176-19187
VerfSemCom: Deep Learning-Driven Verification for Trustworthy Image Semantic Communication
- Open Access
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 5057-5068
Approximate Machine Unlearning through Manifold Representation Forgetting Guided by Self Mode Connectivity
Machine unlearning is a fundamental mechanism that enforces the right to be forgotten. Existing unlearning studies that rely on label manipulation or task-gradient reversal often deliver limited unlearning effectiveness. Moreover, they can undermine the original learning objective and typically do not guarantee equivalence to standard unlearning by retraining. In this paper, we propose ManiF-SMC (Manifold Forgetting with Self Mode Connectivity), motivated by the observation that a model retrained on the remaining data tends to classify erased samples by their semantic similarity to the retained data. We begin with systematically recasting the approximate unlearning as pushing each erased sample away from its original learned manifold representation centroid toward its nearest semantic neighbors in the retained data. This reformulation aligns unlearning with retraining behavior and operates purely in representation space, reducing reliance on labels and task-specific gradients. To tackle the manifold representation-based unlearning problem, ManiF-SMC encapsulates the unlearning and representation preservation goals in a margin-based triplet loss. Because finding a suitable margin for unlearning is challenging, we propose a self-mode-connectivity module that rapidly reconstructs the local manifold to guide the adaptive margin generation for each unlearning case. Extensive experiments on four representative datasets show that ManiF-SMC achieves unlearning effectiveness comparable to state-of-the-art approximate methods while operating solely within the model's representation space.
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 5057-5068
Open AccessApproximate Machine Unlearning through Manifold Representation Forgetting Guided by Self Mode Connectivity
Machine unlearning is a fundamental mechanism that enforces the right to be forgotten. Existing unlearning studies that rely on label manipulation or task-gradient reversal often deliver limited unlearning effectiveness. Moreover, they can undermine the original learning objective and typically do not guarantee equivalence to standard unlearning by retraining. In this paper, we propose ManiF-SMC (Manifold Forgetting with Self Mode Connectivity), motivated by the observation that a model retrained on the remaining data tends to classify erased samples by their semantic similarity to the retained data. We begin with systematically recasting the approximate unlearning as pushing each erased sample away from its original learned manifold representation centroid toward its nearest semantic neighbors in the retained data. This reformulation aligns unlearning with retraining behavior and operates purely in representation space, reducing reliance on labels and task-specific gradients. To tackle the manifold representation-based unlearning problem, ManiF-SMC encapsulates the unlearning and representation preservation goals in a margin-based triplet loss. Because finding a suitable margin for unlearning is challenging, we propose a self-mode-connectivity module that rapidly reconstructs the local manifold to guide the adaptive margin generation for each unlearning case. Extensive experiments on four representative datasets show that ManiF-SMC achieves unlearning effectiveness comparable to state-of-the-art approximate methods while operating solely within the model's representation space.
- 12 May 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 327
ASWmark: A copyright protection approach for audio classification datasets
- 12 May 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 327
ASWmark: A copyright protection approach for audio classification datasets
- 15 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 48 (5) , 5963-5978
BlindU: Blind Machine Unlearning Without Revealing Erasing Data
Machine unlearning enables data holders to remove the contribution of their specified samples from trained models to protect their privacy. However, it is paradoxical that most unlearning methods require the unlearning requesters to first upload their data to the server as a prerequisite for unlearning. These methods are infeasible in many privacy-preserving scenarios where servers are prohibited from accessing users’ data, such as federated learning (FL). In this paper, we explore how to implement unlearning under the condition of not uncovering the erasing data to the server. We propose Blind Unlearning (BlindU), which carries out unlearning using compressed representations instead of original inputs. BlindU only involves the server and the unlearning user: the user locally generates privacy-preserving representations, and the server performs unlearning solely on these representations and their labels. For the FL model training, we employ the information bottleneck (IB) mechanism. The encoder of the IB-based FL model learns representations that distort maximum task-irrelevant information from inputs, allowing FL users to generate compressed representations locally. For effective unlearning using compressed representation, BlindU integrates two dedicated unlearning modules tailored explicitly for IB-based models and uses a multiple gradient descent algorithm to balance forgetting and utility retaining. While IB compression already provides protection for task-irrelevant information of inputs, to further enhance the privacy protection, we introduce a noise-free differential privacy (DP) masking method to deal with the raw erasing data before compressing. Theoretical analysis and extensive experimental results illustrate the superiority of BlindU in privacy protection and unlearning effectiveness compared with the best existing privacy-preserving unlearning benchmarks.
- 15 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 48 (5) , 5963-5978
BlindU: Blind Machine Unlearning Without Revealing Erasing Data
Machine unlearning enables data holders to remove the contribution of their specified samples from trained models to protect their privacy. However, it is paradoxical that most unlearning methods require the unlearning requesters to first upload their data to the server as a prerequisite for unlearning. These methods are infeasible in many privacy-preserving scenarios where servers are prohibited from accessing users’ data, such as federated learning (FL). In this paper, we explore how to implement unlearning under the condition of not uncovering the erasing data to the server. We propose Blind Unlearning (BlindU), which carries out unlearning using compressed representations instead of original inputs. BlindU only involves the server and the unlearning user: the user locally generates privacy-preserving representations, and the server performs unlearning solely on these representations and their labels. For the FL model training, we employ the information bottleneck (IB) mechanism. The encoder of the IB-based FL model learns representations that distort maximum task-irrelevant information from inputs, allowing FL users to generate compressed representations locally. For effective unlearning using compressed representation, BlindU integrates two dedicated unlearning modules tailored explicitly for IB-based models and uses a multiple gradient descent algorithm to balance forgetting and utility retaining. While IB compression already provides protection for task-irrelevant information of inputs, to further enhance the privacy protection, we introduce a noise-free differential privacy (DP) masking method to deal with the raw erasing data before compressing. Theoretical analysis and extensive experimental results illustrate the superiority of BlindU in privacy protection and unlearning effectiveness compared with the best existing privacy-preserving unlearning benchmarks.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dependable and Secure Computing
- Vol. 23 (1) , 1219-1231
SMS: Self-Supervised Model Seeding for Verification of Machine Unlearning
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dependable and Secure Computing
- Vol. 23 (1) , 1219-1231
SMS: Self-Supervised Model Seeding for Verification of Machine Unlearning
- 16 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cognitive Communications and Networking
- Vol. 12, 2131-2145
Information Bottleneck-Based Subgraphs Defending Against Inference Attacks in Federated Graph Learning Systems
Federated Graph learning (FGL) for networked system data enables collaborative model training without sharing raw data, but it also introduces severe privacy risks. Model inversion attacks (MIA), as a type of inference attacks, can extract private graph structure information utilizing the information leakage such as model gradients and nodal attributes. MIA pose significant threats to FGL systems since attackers at the central server can easily access local GNNs’ gradients and potentially infer the private graph structure information of clients. In this paper, we propose a novel scheme to defend against MIA on graph structures in FGL systems while simultaneously preserving their learning performance. Specifically, we utilize the principle of information bottleneck (IB) to extract subgraphs, referred to as IB-subgraphs, that balance the privacy-utility trade-off from the clients’ original subgraphs for local GNN training. IB-subgraphs contain less information regarding the original structure of subgraphs. However, the task-relevant information is retained to a great extent. In particular, we develop a neural network-based approach to address the intractability of estimating the mutual information of graph data during IB optimization. Additionally, we design a subgraph generation algorithm to produce meaningful IB-subgraphs based on the optimization results. Extensive experiments demonstrate that our scheme effectively enhances MIA resilience in FGL while maintaining strong predictive accuracy, making it suitable for secure and intelligent communications and networking applications.
- 16 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cognitive Communications and Networking
- Vol. 12, 2131-2145
Information Bottleneck-Based Subgraphs Defending Against Inference Attacks in Federated Graph Learning Systems
Federated Graph learning (FGL) for networked system data enables collaborative model training without sharing raw data, but it also introduces severe privacy risks. Model inversion attacks (MIA), as a type of inference attacks, can extract private graph structure information utilizing the information leakage such as model gradients and nodal attributes. MIA pose significant threats to FGL systems since attackers at the central server can easily access local GNNs’ gradients and potentially infer the private graph structure information of clients. In this paper, we propose a novel scheme to defend against MIA on graph structures in FGL systems while simultaneously preserving their learning performance. Specifically, we utilize the principle of information bottleneck (IB) to extract subgraphs, referred to as IB-subgraphs, that balance the privacy-utility trade-off from the clients’ original subgraphs for local GNN training. IB-subgraphs contain less information regarding the original structure of subgraphs. However, the task-relevant information is retained to a great extent. In particular, we develop a neural network-based approach to address the intractability of estimating the mutual information of graph data during IB optimization. Additionally, we design a subgraph generation algorithm to produce meaningful IB-subgraphs based on the optimization results. Extensive experiments demonstrate that our scheme effectively enhances MIA resilience in FGL while maintaining strong predictive accuracy, making it suitable for secure and intelligent communications and networking applications.
- 20 June 2025
- Published by Springer Nature
- p. 277-289
Inversion Triplet - A Contrastive Backdoor Mitigation Method for Self-Supervised Vision Encoders
- 20 June 2025
- Published by Springer Nature
- p. 277-289
Inversion Triplet - A Contrastive Backdoor Mitigation Method for Self-Supervised Vision Encoders
- 20 June 2025
- Published by Springer Nature
- p. 3-14
IDIR: Interpolated Diffusion Image Reconstruction for Generalizable Detection of Synthetic Images
- 20 June 2025
- Published by Springer Nature
- p. 3-14
IDIR: Interpolated Diffusion Image Reconstruction for Generalizable Detection of Synthetic Images
- 15 June 2025
- Published by Springer Nature
- p. 271-283
Can Self Supervision Rejuvenate Similarity-Based Link Prediction?
- 15 June 2025
- Published by Springer Nature
- p. 271-283
Can Self Supervision Rejuvenate Similarity-Based Link Prediction?