Yiming Xu

PhD Fellow · Multimodal Document Understanding
Computer Vision Center, Universitat Autònoma de Barcelona

I work in the Vision & Language Group at CVC, UAB, advised by Dimosthenis Karatzas and Ernest Valveny — building agentic document-understanding systems and foundation models for document understanding.

Before Barcelona I earned an MSc at the University of Amsterdam (thesis on Multimodal RAG) and spent two years as principal developer building e-commerce AI agents.

Looking for a research internship or a job in 2027 or 2028, on document agents and agentic RL.

Definition 1 (Document). A page is not an image; it is a tree of meaning that happens to be printed.

News

Nov 2026
Aligning Decoder Attention with Human Fixations is accepted to Findings of EACL 2027.
Jun 2026
MADQA is accepted at ICML 2026 as an Oral.
Apr 2026
AdaNav is accepted at ICDAR 2026 as an Oral.
Jan 2025
Awarded an FPI fellowship (Spanish State Research Agency, AEI). I will do my PhD at the Computer Vision Center within the MuDocU project.
May 2024
KG passage expansion for QA presented at LREC-COLING 2024.

Selected projects

All projects →

Recent publications

All publications →

Recent writing

All writing →
Nov 2026
Notes on Shunyu Yao's talk at TsinghuaShunyu Yao spoke at Tsinghua on 18 November 2026 in a closed session. These are my notes from two write-ups of it, with his remarks in Hong Kong earlier in the month, and what I take from them for work on document agents.
Oct 2025
Planner designHierarchical + leaf embeddings for navigation, not naive top-k retrieval.
Sep 2025
Context engineeringNotes on shaping the context window as a first-class design surface.

Awards

2025
FPI (Formación de Personal Investigador) Fellowship
2017
Chinese Academy of Sciences Science and Innovation Program Scholarship