Devils in Middle Layers of Large Vision-Language Models: Interpreting, Detecting and Mitigating Object Hallucinations via Attention Lens
Zhangqi Jiang, Junkai Chen, Beier Zhu, Tingjin Luo, Yankun Shen, Xu Yang

CVPR, 2025

TL;DR: We address how LVLMs process visual information and whether this process causes hallucination. First, we identify the middle layers are crucial for handling visual data in LVLMs, which can be further divided into two stages: visual information enrichment and semantic refinement. Second, we find that real tokens consistently receive higher attention weights than hallucinated ones, serving as a strong indicator of hallucination. Third, we observe that hallucination tokens often result from attention heads interacting with inconsistent objects.

Deep Incomplete Multi-View Learning Network with Insufficient Label Information
Zhangqi Jiang, Tingjin Luo, Xinyan Liang

AAAI, 2024
[code]

TL;DR: To tackle the double missing of features and labels problem, we propose a novel deep learning model by incorporating graph neural networks and semi-supervised learning into a unified framework.

Pancancer survival prediction using a deep learning architecture with multimodal representation and integration
Ziling Fan$^{\dagger}$, Zhangqi Jiang$^{\dagger}$, Hengyu Liang, Chao Han

Bioinformatics Advances, 2023
[code]

TL;DR: A deep learning architecture to adaptively integrate multi-omics data (e.g., miRNA and CNV) for pancancer survival prediction.