Qiyan Zhao

Institute of Automation, Chinese Academy of Sciences
GitHub  |  Google Scholar  |  Email: qiyanzhao618@gmail.com
Erik Li avatar

About Me

I am a Ph.D. student at the Institute of Automation, Chinese Academy of Sciences, advised by Prof. Xu-Yao Zhang. I work on the mechanistic interpretability of large models and agents, treating internal-mechanism understanding as a first-class design lever rather than a post-hoc diagnostic. My central question is how the internal computations of foundation models give rise to reasoning---and when and why they fail---with the answers directly driving methods for reliable reasoning and reliable agents. I stress-test this programme in VLA and WAM embodied settings, aiming to move interpretable, reliable behaviour from paper to product.

Topics: Interpretable AI, Multimodal Reasoning, Self-Evolving Agents, Embodied Intelligence

Please feel free to reach out if you are interested in related topics.

News

  • 2026.8 — We have 2 papers accepted to EMNLP 2026.
  • 2026.6 — We have 1 paper accepted to IROS 2026.
  • 2026.4 — We have 1 paper accepted to ICIC 2026 Oral.
  • 2026.4 — We have 2 papers accepted to ACL 2026.
  • 2026.2 — We have 1 paper accepted to CVPR 2026.
  • 2026.1 — We have 1 paper accepted to ICRA 2026.
  • 2026.1 — We have 2 paper including 1 oral accepted to ICLR 2026.
  • 2025.10 — We have 1 paper accepted to ACM MM 2025.
  • 2025.9 — We have 1 paper accepted to The Visual Computer.
  • 2025.6 — We have 1 paper accepted to IJCNN 2025.

Selected Publications

C2RoPE paper thumbnail
AdaSGC: Adaptive Suffix Grouping and Caching for Diffusion LMMs
Qiyan Zhao, Guanting Ye, Xu-Yao Zhang, Xiaofeng Zhang, Haoran Lin, Liu Bo, Wenhao Yu, Jiajun Zhang, Da-Han Wang
EMNLP 2026 [Paper] [Code]
C2RoPE paper thumbnail
Context Tokens are Anchors: Understanding the Repeat Curse in dMLLMs from an Information Flow Perspective
Qiyan Zhao, Xiaofeng Zhang, Shuochen Chang, Qianyu Chen, Xiaosong Yuan, Xuhang Chen, Luoqi Liu, Jiajun Zhang, Xu-Yao Zhang, Da-Han Wang
ICLR 2026 [Paper] [Code]
C2RoPE paper thumbnail
MCA-LLaVA: Manhattan Causal Attention for Reducing Hallucination in Large Vision-Language Models
Qiyan Zhao, Xiaofeng Zhang, Yiheng Li, Yun Xing, Xiaosong Yuan, Feilong Tang, Sinan Fan, Xuhang Chen, Da-Han Wang, Xu-Yao Zhang
ACM MM 2025 [Paper] [Code]
C2RoPE paper thumbnail
Hallucination Begins Where Saliency Drops
Xiaofeng Zhang, Yuanchao Zhu, Chaochen Gu, Xiaosong Yuan, Qiyan Zhao, Jiawei Cao, Feilong Tang, Sinan Fan, Yaomin Shen, Chen Shen, Hao Tang
ICLR 2026 Oral [Paper] [Code]
EDAR paper thumbnail
Fixing Semantic Blind Spots in Anchor Tokens of dMLLMs
Ruixuan Xu, Jiexi Xu, Qiyan Zhao, Xiaofeng Zhang
ACL 2026 [Paper]
TokenPenalty paper thumbnail
TokenPenalty: Alleviating Attention Sinks and Positional Decay in LVLMs
Xiaofeng Zhang, Yuanchao Zhu, Qiyan Zhao, Xiaosong Yuan, Jiawei Cao, Xuhang Chen
ACL 2026 [Paper]
SoPE paper thumbnail
SoPE: Spherical Coordinate-Based Positional Embedding for Enhancing Spatial Perception of 3D LVLMs
Koon-Ting Yip, Qiyan Zhao, Wenhao Yu, Liangyu Yuan, Mingkai Li, Xiaofeng Zhang, Jianmin Ji, Yanyong Zhang, Qing Jiang, Ka-Veng Yuen
CVPR 2026 [Paper]
C2RoPE paper thumbnail
C2ROPE: Causal Continuous Rotary Positional Encoding for 3D Large Multimodal Models Reasoning
Koon-Ting Yip, Qiyan Zhao, Wenhao Yu, Xiaofeng Zhang, Jianming Ji, Yanyong Zhang, Ka-Veng Yuen
ICRA 2026 [Paper] [Code]
C2RoPE paper thumbnail
Embodied Chain-of-Thought Model via Interleaved Reasoning
Bo Liu, Qiyan Zhao, Liming Zhang, Yuhan Wang, Yong Liu
IROS 2026 [Paper] [Code]

Services