Hi! ๐Ÿฅฐ Iโ€™m Sen Nie (่‚ๆฃฎ), a first-year Masterโ€™s student at the Visual Information Processing and Learning (VIPL) group, Institute of Computing Technology (ICT), Chinese Academy of Sciences (CAS). I also conduct research at the State Key Laboratory of AI Safety, under the supervision of Profs. Jie Zhang, Shiguang Shan (IEEE Fellow), and Xilin Chen (IEEE Fellow).

My research is currently focused on the frontier of AI Safety. Iโ€™m passionate about building robust and trustworthy AI systems and am always open to collaboration or a good academic chat!

๐Ÿ”ฅ News

  • May 2026 ๐ŸŽ‰ Paper CSR has been accepted by ICML 2026! Thanks to all co-authors!
  • Mar 2026 ๐ŸŽ‰ Our work BIMM has been accepted by TCSVT 2026! Thanks to all co-authors!
  • Feb 2026 ๐ŸŽ‰ Paper V-Attack is accepted by CVPR 2026! Thanks to all co-authors!
  • Jan 2026 ๐Ÿฅ‡ Our team AIST_ICT achieved 1st place in Anti-BAD (IEEE SaTML 2026)! Congratulations to all team members!
  • Jun 2025 ๐ŸŽ“ Graduated from Wuhan University; awarded Outstanding Graduate and Outstanding Undergraduate Thesis honors.

๐Ÿ“ Publications

Contrastive Spectral Rectification: Test-Time Defense towards Zero-shot Adversarial Robustness of CLIP Code
Sen Nie, Jie Zhang, Zhuo Wang, Shiguang Shan, Xilin Chen
International Conference on Machine Learning (ICML), 2026
Contrastive Spectral Rectification (CSR) is an input-adaptive test-time defense for VLMs that optimizes a spectral-guided contrastive rectification perturbation to realign each input with the natural image manifold.
BIMM: Brain Inspired Masked Modeling for Video Representation Learning
Jie Zhang, Zhifan Wan, Sen Nie, Changzhen Li, Shiguang Shan, Xilin Chen
IEEE Transactions on Circuits and Systems for Video Technology (TCSVT), 2026
Brain Inspired Masked Modeling (BIMM) learns video representations with paired ventral image and dorsal video ViT branches trained by masked modeling, progressive intermediate-target reconstruction, and partial parameter sharing to model layered object and motion processing in the human visual system.
V-Attack: Targeting Disentangled Value Features for Controllable Adversarial Attacks on LVLMs Code
Sen Nie, Jie Zhang, Jianxin Yan, Shiguang Shan, Xilin Chen
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026
V-Attack precisely manipulates image semantics in LVLMs by attacking transformer value features rather than entangled patch tokens, using Self-Value Enhancement to strengthen local semantic information and Text-Guided Value Manipulation to optimize a specified source concept toward a target concept.

๐Ÿ† Awards

๐ŸŽ– Honors

  • At WHU, awarded Outstanding Graduate (2025), Outstanding Undergraduate Thesis (2025), and Outstanding Student Award (2021-2024)
  • At WHU, awarded Lei Jun Scholarship (Top 1%), NSFOCUS (LvMeng) Scholarship (Top 0.5%), and First-Class Scholarship *3 (Top 5%)

๐Ÿ“– Education

Institute of Computing Technology, Chinese Academy of Sciences
Beijing, China
Sep. 2025 - Present
M.S. Candidate in Computer Science and Technology
Wuhan University
Wuhan, China
Sep. 2021 - Jul. 2025
B.Eng. in Cyberspace Security