News
Updates
2026
Joined Kuaishou as a Research Intern.
CoTA++, the extended version of our ICLR 2026 paper, is on arXiv. It traces the repeat curse in diffusion MLLMs to three information-flow disruptions and fixes each one without training.
Context-Aware Cluster Decoding, on semantic anchors for coherent diffusion MLLMs, is accepted to the EMNLP 2026 main conference.
New preprint: when do the skills of self-evolving agents actually help? Coalition-Aware Skill Reliability is on arXiv.
Embodied chain-of-thought via interleaved reasoning is accepted to IROS 2026 as an Oral.
Two papers, on attention sinks in LVLMs and on anchor tokens in diffusion MLLMs, are accepted to Findings of ACL 2026.
One paper on efficient vision–language models is accepted to ICIC 2026 as an Oral.
Started my Ph.D. at the Institute of Automation, Chinese Academy of Sciences, advised by Prof. Xu-Yao Zhang.
SoPE, a spherical-coordinate positional embedding for 3D LVLMs, is accepted to CVPR 2026.
C²RoPE, a causal continuous rotary positional encoding for 3D LMM reasoning, is accepted to ICRA 2026.
Two papers are accepted to ICLR 2026, including Hallucination Begins Where Saliency Drops as an Oral.
2025
MCA-LLaVA, on reducing hallucination in LVLMs with Manhattan causal attention, is accepted to ACM MM 2025.
Few-shot cross-modal text detection via CLIP is accepted to The Visual Computer.
TextMamba, a scene text detector built on Mamba, is accepted to IJCNN 2025.