Tao Gong
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Tao Gong has a total of 62 co-authors. Here are the top co-authors by number of shared publications:
- Qi Chu
University of Science and Technology of China
21 shared publications - Nenghai Yu
University of Science and Technology of China, Hefei, China
16 shared publications - Bin Liu
University of Science and Technology of China, Hefei, China
10 shared publications - Changtao Miao
University of Science and Technology of China, Hefei, China
9 shared publications Yue Wu
Alibaba Group, Alibaba Cloud, Hangzhou, China
5 shared publications- Zhentao Tan
School of Cyber Science and Technology, University of Science and Technology of China, Hefei, China
5 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 16926-16933
DDL: A Large-Scale Dataset for Deepfake Detection and Localization in Diversified Real-World Scenarios
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 16926-16933
Open AccessDDL: A Large-Scale Dataset for Deepfake Detection and Localization in Diversified Real-World Scenarios
- 18 September 2026
- Published by Springer Nature
- p. 401-419
ANFI: Rethinking Neighbor Feature Interaction in Person Re-ID
- 18 September 2026
- Published by Springer Nature
- p. 401-419
ANFI: Rethinking Neighbor Feature Interaction in Person Re-ID
- 12 September 2026
- Published by Springer Nature
- p. 358-375
EraseSAE: Surgical Concept Erasure in Text-to-Video Diffusion Models via Sparse Autoencoders
- 12 September 2026
- Published by Springer Nature
- p. 358-375
EraseSAE: Surgical Concept Erasure in Text-to-Video Diffusion Models via Sparse Autoencoders
- 18 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21 (15566013) , 5149-5164
PVDI: Preserving Vital and Disrupting Irrelevant Latent Attentions for Robust Backdoor Defense
Backdoor attacks pose a critical threat to the security and reliability of deep neural networks (DNNs), enabling malicious triggers to manipulate model predictions while maintaining normal performance on clean inputs. Addressing this challenge requires robust defense mechanisms that eliminate backdoors without compromising model utility. This paper introduces PVDI (Preserving Vital and Disrupting Irrelevant attentions), a novel blind purification method designed to neutralize backdoors while retaining the primary prediction capabilities of both backdoored and clean models. PVDI leverages vital and irrelevant attention mechanisms during fine-tuning: vital attention is preserved to maintain the model’s core functionality, while irrelevant attention is disrupted to neutralize backdoor behavior. Extensive evaluations across diverse datasets and attack scenarios demonstrate PVDI’s superior purification performance, achieving significant reductions in attack success rates while preserving the utility of backdoored models and minimizing adverse impact on clean models. PVDI outperforms existing state-of-the-art defenses and sets a new benchmark in backdoor defense. This work represents a significant step forward in combating backdoor attacks. Our code will be made publicly available upon publication.
- 18 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21 (15566013) , 5149-5164
PVDI: Preserving Vital and Disrupting Irrelevant Latent Attentions for Robust Backdoor Defense
Backdoor attacks pose a critical threat to the security and reliability of deep neural networks (DNNs), enabling malicious triggers to manipulate model predictions while maintaining normal performance on clean inputs. Addressing this challenge requires robust defense mechanisms that eliminate backdoors without compromising model utility. This paper introduces PVDI (Preserving Vital and Disrupting Irrelevant attentions), a novel blind purification method designed to neutralize backdoors while retaining the primary prediction capabilities of both backdoored and clean models. PVDI leverages vital and irrelevant attention mechanisms during fine-tuning: vital attention is preserved to maintain the model’s core functionality, while irrelevant attention is disrupted to neutralize backdoor behavior. Extensive evaluations across diverse datasets and attack scenarios demonstrate PVDI’s superior purification performance, achieving significant reductions in attack success rates while preserving the utility of backdoored models and minimizing adverse impact on clean models. PVDI outperforms existing state-of-the-art defenses and sets a new benchmark in backdoor defense. This work represents a significant step forward in combating backdoor attacks. Our code will be made publicly available upon publication.
- 16 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21 (15566013) , 5455-5468
ABDP: Adversarial Backdoor Detection and Purification
In domains like driving and healthcare, deep learning models often rely on large, diverse datasets that can inadvertently harbor backdoor attacks. In this paper, we propose ABDP, a novel post-processing defense method, to effectively remove backdoor contamination from datasets and generate clean models without relying on any pre-existing clean data. ABDP capitalizes on the intrinsic link between untargeted adversarial attacks and backdoor attacks to detect the presence of backdoor attacks within trained models and ascertain their target labels. Subsequently, it trains a clean model capable of recognizing all labels except the target label, thus treating poisoned data as in-distribution and clean data of the target label as out-of-distribution. This distinction enables the identification of backdoor poisoned data. Finally, ABDP applies unlearning techniques to effectively eradicate the backdoor from the model. Extensive experimental evaluations across diverse datasets and against multiple backdoor attack scenarios validate the robustness and state-of-the-art performance of our approach. By employing ABDP for data and model cleansing, the attack success rate of resulting models is reduced to 1% or less, while retaining approximately 70% or more clean data at a true positive rate of 0.01 false positive rate. Notably, ABDP exhibits no adverse impact when applied to purely clean datasets, owing to its ability to detect backdoor presence in models before cleansing. Thus, our proposed method achieves state-of-the-art performance in cleansing both backdoor-poisoned data and backdoor models. The code for ABDP will be made available upon publication of the paper.
- 16 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21 (15566013) , 5455-5468
ABDP: Adversarial Backdoor Detection and Purification
In domains like driving and healthcare, deep learning models often rely on large, diverse datasets that can inadvertently harbor backdoor attacks. In this paper, we propose ABDP, a novel post-processing defense method, to effectively remove backdoor contamination from datasets and generate clean models without relying on any pre-existing clean data. ABDP capitalizes on the intrinsic link between untargeted adversarial attacks and backdoor attacks to detect the presence of backdoor attacks within trained models and ascertain their target labels. Subsequently, it trains a clean model capable of recognizing all labels except the target label, thus treating poisoned data as in-distribution and clean data of the target label as out-of-distribution. This distinction enables the identification of backdoor poisoned data. Finally, ABDP applies unlearning techniques to effectively eradicate the backdoor from the model. Extensive experimental evaluations across diverse datasets and against multiple backdoor attack scenarios validate the robustness and state-of-the-art performance of our approach. By employing ABDP for data and model cleansing, the attack success rate of resulting models is reduced to 1% or less, while retaining approximately 70% or more clean data at a true positive rate of 0.01 false positive rate. Notably, ABDP exhibits no adverse impact when applied to purely clean datasets, owing to its ability to detect backdoor presence in models before cleansing. Thus, our proposed method achieves state-of-the-art performance in cleansing both backdoor-poisoned data and backdoor models. The code for ABDP will be made available upon publication of the paper.
- 1 November 2025
- Published by Springer Nature
- p. 301-312
Exploring Generalized Features For LLM-Generated Text Detection
- 1 November 2025
- Published by Springer Nature
- p. 301-312
Exploring Generalized Features For LLM-Generated Text Detection
- 1 November 2025
- Published by Springer Nature
- p. 289-300
Exploiting Feature Gating and Injection For Multi-modal Manipulation Detection and Grounding
- 1 November 2025
- Published by Springer Nature
- p. 289-300
Exploiting Feature Gating and Injection For Multi-modal Manipulation Detection and Grounding
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 5431-5440
Towards Good Generalizations for Diffusion Generated Image Detection Using Multiple Reconstruction Contrastive Learning
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 5431-5440
Open AccessTowards Good Generalizations for Diffusion Generated Image Detection Using Multiple Reconstruction Contrastive Learning
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 3903-3912
Mixture-of-Noises Enhanced Forgery-Aware Predictor for Multi-Face Manipulation Detection and Localization
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 3903-3912
Open AccessMixture-of-Noises Enhanced Forgery-Aware Predictor for Multi-Face Manipulation Detection and Localization
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 13235-13242
MFFI: Multi-Dimensional Face Forgery Image Dataset for Real-World Scenarios
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 13235-13242
Open AccessMFFI: Multi-Dimensional Face Forgery Image Dataset for Real-World Scenarios