I am a senior undergraduate student in Computer Science and Engineering at Southeast University (SEU). In 2026, I will join the Institute of Automation, Chinese Academy of Sciences (CASIA) as a Ph.D. student, under the supervision of Prof. Tieniu Tan.

My primary research focus spans Multimodal Large Language Models (MLLMs), Agentic AI, and Trustworthy AI. I have published several papers at top-tier AI conferences such as NeurIPS, ICLR, ICML, ICCV, and ACL.

I am always open to and excited about various types of research collaborations! Whether you are interested in discussing MLLMs, exploring new paradigms for AI agents, or working on trustworthy AI, please feel free to reach out.

πŸ“§ Email: qianshanwei7@gmail.com

πŸ“’ News

  • 2026.05: πŸ† β€œA-MemGuard: A proactive defense framework for llm-based agent memory” accepted by ICML 2026. See you in Seoul, Korea! Media Report (ζ–°ζ™Ίε…ƒ)
  • 2026.03: πŸ† Our paper β€œTime Is All It Takes: Spike-Retiming Attacks on Event-Driven Spiking Neural Networks” accepted by ICLR 2026!
  • 2026.03: πŸ† Our work on Selective Concept Unlearning (SCU) for segmentation foundation models accepted by AAAI 2026.
  • 2026.01: Our survey/paper β€œToward Efficient Agents” is out on arXiv.
  • 2025.07: πŸ† β€œForget the Token and Pixel: Rethinking Gradient Ascent for Concept Unlearning” accepted by ACL 2025.
  • 2025.05: πŸ† β€œScaling large motion models with million-level human motions” accepted by ICML 2025, with media coverage by 量子位 (QbitAI).
  • 2025.01: πŸ† β€œMotionCtrl: A Real-time Controllable Vision-Language-Motion Model” accepted by ICCV 2025.
  • 2024.10: πŸ† β€œSingle image unlearning: Efficient machine unlearning in MLLMs” accepted by NeurIPS 2024.

πŸ“š Selected Publications

(*: Equal contribution)

ICLR 2026
sym

Time Is All It Takes: Spike-Retiming Attacks on Event-Driven Spiking Neural Networks
Yi Yu, Q Zhang, S Ye, X Lin, Qianshan Wei, K Wang, W Yang, D Tao, X Jiang
International Conference on Learning Representations (ICLR), 2026

AAAI 2026
sym

Forget What Has Seen: Selective Concept Unlearning in Segmentation Foundation Models
Miaozeng Du, Jiaqi Li, Sirui Pan, Yi Zhan, Guilin Qi, Yuxin Zhang, Rihui Jin, Yinjia Shu, Qianshan Wei
Proceedings of the AAAI Conference on Artificial Intelligence (AAAI), 2026

ICML 2026
sym

A-MemGuard: A proactive defense framework for llm-based agent memory
Qianshan Wei, T Yang, Y Wang, X Li, L Li, Z Yin, Y Zhan, T Holz, Zhiqiang Lin, XiaoFeng Wang
International Conference on Machine Learning (ICML), 2026 Media Report (ζ–°ζ™Ίε…ƒ)

ICML 2025
sym

Scaling large motion models with million-level human motions
Y Wang, S Zheng, B Cao, Qianshan Wei, W Zeng, Q Jin, Z Lu

International Conference on Machine Learning (ICML), 2025 Media Report (量子位)

ICCV 2025
sym

MotionCtrl: A Real-time Controllable Vision-Language-Motion Model
B Cao, S Zheng, Y Wang, L Xia, Qianshan Wei, Q Jin, J Liu, Z Lu

International Conference on Computer Vision (ICCV), 2025

ACL 2025
sym

Forget the Token and Pixel: Rethinking Gradient Ascent for Concept Unlearning in Multimodal Generative Models
J Li, C Zhang, M Du, H Zhang, Y Chen, Qianshan Wei, J Fang, R Wang, S Bi, G Qi

Findings of the Association for Computational Linguistics (ACL), 2025

NeurIPS 2024
sym

Single image unlearning: Efficient machine unlearning in multimodal large language models
Jiaqi Li*, Qianshan Wei*, C Zhang, G Qi, M Du, Y Chen, S Bi, F Liu

Advances in Neural Information Processing Systems (NeurIPS) 37, 2024

πŸ“š Recent Work & Preprints

arXiv 2025
sym

Dual-Priv Pruning: Efficient Differential Private Fine-Tuning in Multimodal Large Language Models Qianshan Wei,
Jiaqi Li, Zihan You, Yi Zhan, Kecen Li, Jialin Wu, Xinfeng Li, Hengjun Liu, Yi Yu, Bin Cao, Y Xu, Y Liu, Guilin Qi.

arXiv preprint, 2025 > πŸ’‘ Note: This work received all positive reviews (5, 4, 4, 4) from NeurIPS 2025 Main Track reviewers and another strong but unsuccessful round at ICML 2026 with scores 5, 4, 4, 4. We are still looking for the right home for this contribution.

arXiv 2026
sym

Toward Efficient Agents: Memory, Tool learning, and Planning
X Yang, L Li, H Zhou, T Zhu, X Qu, Y Fan, Qianshan Wei, R Ye, L Kang, Y Qin, et al.
arXiv preprint, 2026

  • Being-m0.5: A real-time controllable vision-language-motion model,
  • B Cao, Qianshan Wei et al., arXiv 2025.

πŸ“ Professional Service

  • Reviewer: CVPR, ICML, NeurIPS

πŸŽ“ Educations

  • Ph.D. in Pattern Recognition and Intelligent Systems (Incoming), Institute of Automation, Chinese Academy of Sciences (CASIA), 2026–Present
  • B.Eng. in Computer Science and Engineering, Southeast University (SEU), 2022–2026