AI researcher
Yuyang Dong (David Dong)
Chief Research Scientist, SB Intuitions (SoftBank Group)
The University of Osaka · Invited Associate Professor
I connect data, models, and agents in a continuous learning loop, using feedback from real-world applications to optimize the whole system.
My work on Japanese foundation models spans Sarashina3 at SB Intuitions (SoftBank Group) and cotomi / cotomi-v2 at NEC.
My research covers pretraining, post-training, and model evaluation.
SB Intuitions · SoftBank Group
Japanese foundation model research and development.
Official technical blogNEC
Core team member in NEC’s Japanese LLM development.
NEC interviewJellyfish applies instruction-tuned language models to data preprocessing. DeepJoin discovers joinable tables in data lakes.
SCAN connects document layout analysis with textual and visual retrieval. My work also includes chart and diagram understanding at NEC.
Rei Taniguchi, Yuyang Dong, Makoto Onizuka, Chuan Xiao
Nobuhiro Ueda*, Yuyang Dong*, Krisztián Boros, Daiki Ito, Takuya Sera, Masafumi Oyamada
Yahan Yu, Yuyang Dong , Masafumi Oyamada
Haochen Zhang*, Yuyang Dong*, Chuan Xiao, Masafumi Oyamada
* Equal contribution.
Pretraining, reinforcement learning, and agent capabilities. Published on the SB Intuitions technical blog.
Slides from 最強DB講義 #37.
Training data, model design, and evaluation, discussed with the development team.
Invited Associate Professor
Chief Research Scientist
Sarashina model development
Principal Researcher, previously Special Researcher and Researcher.
Core team member, cotomi / cotomi-v2; chart and document understanding.
Master of Engineering & Ph.D. in Computer Science
Bachelor’s degree in Computer Science
