Qibing Ren「任麒冰」I am a PhD candidate at the DMCV lab of Shanghai Jiao Tong University. I am currently the class leader of Wu Wenjun Honorary Doctoral Program at SJTU. My PhD advisor is Lizhuang Ma. Previously, I received my B.S. from SJTU. I also spent time at AI research labs of MiniMax, Tencent, ByteDance and Shanghai AI Lab. Email: renqibing [at] sjtu.edu.cn |
|
Research TopicsI'd like to train superintelligent AI that is both 'smart' and 'safe' for human goodness. So, to make safe AI, I am interested in human-computer interaction, AI alignment, and AI safety. I am also developing advanced multimodal AI systems. |
Publications |
LLMs know their vulnerabilities: Uncover Safety Gaps through Natural Distribution ShiftsFirst author Association for Computational Linguistics (ACL), 2025 arxiv / code / ACL 2025 Outstanding Paper Award (top 0.3%) |
When Autonomy Goes Rogue: Preparing for Risks of Multi-Agent Collusion in Social SystemsFirst author arXiv, 2025 arxiv / code / website / |
Frontier AI Risk Management Framework in Practice: A Risk Analysis Technical ReportTeam work at Shanghai AI Lab, Leader of the Collusion Risk Section. Technical Report, 2025 arxiv / website / blog / Featured by Jack Clark (Anthropic Co-founder) |
MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning AttentionTeam work at MiniMax arXiv, 2025 arxiv / code / website / The world’s first open-weight, large-scale hybrid-attention reasoning model. |
One RL to See Them All: Visual Triple Unified Reinforcement LearningTeam work at MiniMax arXiv, 2025 arxiv / code / website / V-Triune is the first system to jointly advancing visual reasoning and perception capabilities of VLMs within a single RL pipeline. |
Certified Robustness on Visual Graph Matching via Searching Optimal Smoothing RangeFourth author ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), 2024 arxiv / |
Codeattack: Revealing safety generalization challenges of large language models via code completionFirst author Association for Computational Linguistics (ACL), 2024 arxiv / code / |
ROCO: A General Framework for Evaluating Robustness of Combinatorial Optimization Solvers on GraphsFourth author International Conference on Learning Representations (ICLR), 2023 arxiv / code / |
Rethinking and Improving Robustness of Convolutional Neural Networks: a Shapley Value-based Approach in Frequency DomainSecond author Neural Information Processing Systems (NeurIPS), 2022 arxiv / code / NeurIPS Spotlight 2022 |
DICE: Domain-attack Invariant Causal Learning for Improved Data Privacy Protection and Adversarial RobustnessFirst author ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), 2022 arxiv / code / |
Appearance and Structure Aware Robust Deep Visual Graph Matching: Attack, Defense and BeyondFirst author IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022 arxiv / code / |
Awards
|
Yuanqing Yang Scholarship (4 Ph.D. candidate fellows in CS), SJTU. 2025 |
Professional Services
|
PC member/Reviewer: TPAMI, ICML'21-25, NeurIPS'21-25, ICLR'22-26, AAAI'21-25, AISTATS'24-26 |
|
Design and source code from Jon Barron's website |