I am a master's student in the Graduate School of Science and Engineering at the University of Toyama, pursuing a Master of Mathematics and Informatics (expected March 2027). I'm interested in representation learning for visual understanding and prediction, particularly how the organization of a representation space shapes what a model can learn and how its features can be used across tasks.
My research spans vision-language dataset distillation, 3D human-object interaction generation, and self-supervised visual representation learning. I am currently studying how relationships among patch representations evolve during training, with a focus on the causes of structural collapse and dense feature degradation. My goal is to learn representations in which heterogeneous modalities complement one another, preserving useful spatial and temporal details while reducing computational overhead.
Research Interests
Representation Learning
Organizing visual and multimodal information into representations that support understanding and prediction across tasks.
Self-Supervised Learning
Studying how relationships among features evolve during training and which structures support useful visual representations.
Multimodal Learning
Learning how heterogeneous modalities can complement one another in shared representations for understanding and prediction.
News
View allI was recommended for Tuition Fee Exemption for Outstanding Students at the University of Toyama.
My paper has been accepted to PRICAI 2026.
- Yaoting Huang, Wenqi Cai, Yijin Wei, Katsuya Hotta, Chunzhi Gu, Haoran Xie and Chao Zhang, TriP-AD: Logical Anomaly Detection under Vision-Induced Distractions via Tri-Path Normality Scoring
2025年 電気学会 電子・情報・システム部門大会 にてポスター発表を行い、優秀ポスター賞 を受賞しました。
- Wenqi Cai, Yawen Zou, Chao Zhang, Dataset Distillation with Foreground Extraction
