Mitigating Hallucination of Large Vision-Language Models via Dynamic Logits Calibration
A training-free framework for reducing hallucination in vision-language models through dynamic logits calibration.
Jiahe Chen
•
•
2 min readAbout Me
I am a fourth-year (senior) undergraduate student at Zhejiang University of Technology, and an incoming Direct PhD student at Zhejiang University (ZJU) starting in 2026. Currently, I am enrolled in a joint training program at Shanghai Innovation Institute (SII), co-advised by Prof. Yongliang Shen (ZJU) and Prof. Xipeng Qiu (SII).
My research focuses on NLP and LLMs, particularly Autonomous Agents and Multimodal Intelligence, aiming to build intelligent systems that can perceive, reason, and act in complex open-world environments. My undergraduate supervisor is Prof. Jianwei Zheng. Previously, I gained research experience across medical image processing, knowledge graphs, and vision-language models, and I have worked closely with Jiahe Chen (my senior colleague). I am always open to collaborations and connections in the research community.
Education
Direct PhD Student (Incoming)
Zhejiang University & Shanghai Innovation Institute
Computer Science and Technology (Undergraduate)
Zhejiang University of Technology
Research Interests
Large Language Models
Natural Language Processing
Autonomous Agents
Multimodal Learning
🔥 Recent News
- 2025.07 - One paper is accepted by Bioinformatics.
- 2025.03 - One paper is accepted by ICME 2025.
- 2024.12 - I was honored to be awarded the National Scholarship.
📝 All Publications
(2025).
Mitigating Hallucination of Large Vision-Language Models via Dynamic Logits Calibration.
arXiv.
(2025).
BioGraphFusion: Graph Knowledge Embedding for Biological Completion and Reasoning.
Bioinformatics.
🔬 All Research
Read more
BioGraphFusion: Graph Knowledge Embedding for Biological Completion and Reasoning
A framework fusing tensor decomposition with graph structure propagation for biological knowledge graph completion and reasoning.
Yitong Lin
•
•
1 min read
C3S3: Complementary Competition and Contrastive Selection for Semi-Supervised Medical Image Segmentation
A semi-supervised medical image segmentation model with dynamic competition and contrastive learning mechanisms.
