TranscEngram

Leadership team

Our team brings together experience in intelligence research, engineering, commercialization and finance to build systems for the physical world.

Portrait of Yi Ma

Founder

Yi Ma

Yi Ma studies 3D vision and representation learning. He co-developed the principle of maximal coding rate reduction (MCR²). His collaborations with Yann LeCun and colleagues on vision-language models and decision learning have appeared at CVPR and NeurIPS.

Ma graduated from Tsinghua University and earned his PhD in electrical engineering and computer sciences at UC Berkeley. He later held a professorship at Berkeley, served as executive dean of ShanghaiTech’s School of Information Science and Technology, and led the Visual Computing Group at Microsoft Research Asia.

  • Founding Director of the School of Computing and Data Science and Chair Professor of Artificial Intelligence at HKU.
  • Shared the 1999 ICCV David Marr Prize with his co-authors for their work on 3D reconstruction.
  • Fellow of IEEE, ACM and SIAM.
Portrait of Zhiru Shi

Zhiru Shi holds a Ph.D. from the University of Surrey. He was a postdoctoral researcher at Shanghai Jiao Tong University and a research assistant professor at ShanghaiTech University, working on 3D vision, light-field imaging and video coding.

He founded YOKE Intelligence in 2018 and later served as Vice President of DGene, bringing 3D vision and digital twin systems into practical use. He is co-founder and CEO of TranscEngram, where he leads commercialization and fundraising.

Portrait of Shenghua Gao

CSO

Shenghua Gao

Personal website

Shenghua Gao is a tenured associate professor at HKU’s School of Computing and Data Science. His research covers 3D and 4D reconstruction, generative models and embodied AI. He earned his bachelor’s degree at USTC and PhD at Nanyang Technological University, and was previously a professor at ShanghaiTech University.

He received a Microsoft Research Fellowship and the ACM Shanghai Young Research Scientist Award, and served as publicity chair for CVPR 2024 and area chair for leading conferences. He is the corresponding author of CUPID, which jointly recovers object shape, texture and camera pose from a single image.

Portrait of Yanchao Yang

CTO

Yanchao Yang

Personal website

Yanchao Yang is an assistant professor in HKU’s School of Computing and Data Science and Department of Electrical and Computer Engineering. He leads the InfoBodied AI Lab, studying 3D perception, action concept learning and robot manipulation with less reliance on human annotation.

He earned his PhD in computer science at UCLA with Stefano Soatto and conducted postdoctoral research at Stanford with Leonidas Guibas. His honors include the RGC Early Career Award and Meta Research Award. He has served as an area chair for NeurIPS, ICLR and RSS.

Portrait of Chaobing Song

CPO

Chaobing Song

Personal website

Chaobing Song holds a Ph.D. from Tsinghua University. He conducted research at UC Berkeley with Yi Ma and was a postdoctoral researcher at the University of Wisconsin–Madison. He works on large-scale optimization and machine learning and co-authored the MCR² paper.

At Huawei, he led projects in retrieval, attention operator fusion and reinforcement learning post-training. At TranscEngram, he oversees training systems, post-training, model updates and adaptation to domestically developed computing hardware.

Portrait of Bei Yu

CFO

Bei Yu

Bei Yu holds a bachelor’s degree in business administration from Southwestern University of Finance and Economics. She served as deputy branch head at Standard Chartered and HSBC, with responsibilities across finance, legal and compliance, operations and organizational management.

She co-founded TranscEngram in 2023 and serves as CFO. Her experience includes financing structures, financial modeling, internal controls and operational management.

TranscEngram Research

Foundations for autonomous intelligence, from visual representations and 3D reconstruction to robot learning.

Explore TranscEngram Research

ICML 2025

SimDINO

SimDINO simplifies self-supervised learning to build transferable visual representations for machine perception and understanding.

CVPR 2026 · Highlight

CUPID

CUPID estimates an object’s 3D shape, texture and camera pose from a single image. Yi Ma is a co-author and Shenghua Gao is the corresponding author.

This four-level benchmark tests whether robots can still achieve a demonstrated goal when the layout or available tools change. Yanchao Yang is the corresponding author.

2026 · arXiv

YOCO

Calibrates hand-motion capture from a small set of paired poses, reducing operator and device bias in dexterous teleoperation without fine-tuning a separate model for each session.

ECCV 2026

Pointer-CAD v2

Separates design planning from geometric construction to generate CAD models with explicit dimensions. Continuous parameters preserve dimensional precision and support editing through the design plan.

2026 · arXiv

RoboLineage

Links robot rollout evidence, dataset revisions, policy training and evaluation in a traceable history. Failure reviews guide the next collection round and support continued policy improvement.