Xin Liu

Xin Liu

Auto-generated profile, based on publicly available information

Xin Liu

School of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan, China

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
24 January 2025 in IEEE Transactions on Information Forensics and Security
Verified
Journal Article
Closed access
XLXin Liu
Xin Liu
YYYichen Yang
Yichen Yang
KHKun He
Kun He
JHJohn E. Hopcroft
John E. Hopcroft
Though deep neural networks exhibit superior performance on various tasks, they are still plagued by adversarial examples. Adversarial training has been demonstrated to be the most effective method to defend against adversarial attacks. However, existing adversarial training methods show that the model robustness has apparent oscillations and overfitting issues in the training process, degrading the defense efficacy. To address these issues, we propose a novel framework called Parameter Interpolation Adversarial Training (PIAT). PIAT tunes the model parameters between each epoch by interpolating the parameters of the previous and current epochs. It makes the decision boundary of model change more moderate and alleviates the overfitting issue, helping the model converge better and achieving higher model robustness. In addition, we suggest using the Normalized Mean Square Error (NMSE) to further improve the robustness by aligning the relative magnitude of logits between clean and adversarial examples rather than the absolute magnitude. Extensive experiments conducted on several benchmark datasets demonstrate that our framework could prominently improve the robustness of both Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs).

1 January 2026 in IEEE Transactions on Emerging Topics in Computational Intelligence
Verified
Journal Article
Closed access
RDRenhua Ding
Renhua Ding
AZAoyang Zhou
Aoyang Zhou
XLXin Liu
Xin Liu
WWWei Wang
Wei Wang
KHKun He
Kun He

16 October 2024
Verified
Book Chapter
Open access
PDF file
AZAoyang Zhou
Aoyang Zhou
YKYasmeen M. Khedr
Yasmeen M. Khedr
XLXin Liu
Xin Liu
KHKun He
Kun He
Deep Neural Networks (DNNs) have been widely used in Synthetic Aperture Radar Automatic Target Recognition (SAR-ATR) to achieve better performance on SAR image recognition. However, DNNs are known to be vulnerable to adversarial examples, crafted by injecting imperceptible perturbations into clean examples. There is limited research on adversarial attacks for SAR-ATR and most of them focus on white-box attacks, which are impractical in real-world scenarios. Existing works have shown that gradient-based transferable attacks, initially designed for optical images, can effectively enhance the black-box attack performance of adversarial examples. However, we observe that they have rather limited efficacy on SAR images. In this work, we propose a novel gradient-based optimization attack called Sampling-based TEacher Guided (STEG) attack to improve the transferability of SAR adversarial examples. STEG leverages a teacher adversarial example to guide the generation of a student adversarial example by utilizing the ultimate perturbation direction. It stabilizes the update direction by simultaneously incorporating the gradients of both teacher and student adversarial examples during the attack iteration process. Moreover, STEG introduces gradually diminishing random noise to further enhance the attack transferability. Extensive experiments conducted on the MSTAR dataset demonstrate that STEG yields significantly higher performance than other advanced gradient-based transferable attacks in both black-box and white-box settings.

2 November 2025 in arXiv
Verified
Preprint
Open access
PDF file
XLXin Liu
Xin Liu
YYYichen Yang
Yichen Yang
KHKun He
Kun He
JHJohn E. Hopcroft
John E. Hopcroft
Though deep neural networks exhibit superior performance on various tasks, they are still plagued by adversarial examples. Adversarial training has been demonstrated to be the most effective method to defend against adversarial attacks. However, existing adversarial training methods show that the model robustness has apparent oscillations and overfitting issues in the training process, degrading the defense efficacy. To address these issues, we propose a novel framework called Parameter Interpolation Adversarial Training (PIAT). PIAT tunes the model parameters between each epoch by interpolating the parameters of the previous and current epochs. It makes the decision boundary of model change more moderate and alleviates the overfitting issue, helping the model converge better and achieving higher model robustness. In addition, we suggest using the Normalized Mean Square Error (NMSE) to further improve the robustness by aligning the relative magnitude of logits between clean and adversarial examples rather than the absolute magnitude. Extensive experiments conducted on several benchmark datasets demonstrate that our framework could prominently improve the robustness of both Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs).