Tony Yang

I'm a Marie Curie Fellow pursuing Ph.D. at IMDEA Networks Institute supervised by Prof. Dr. Joerg Widmer. I conduct research on efficient multimodal wireless sensing systems for human and obstacle detection and tracking, as part of the MSCA 6th Sense Project.

Prior, I have also worked as a full-time AI research engineer at ImPhys TU Delft with Dr. Tao Qian, where I optimized AI models for medical imaging through advanced pruning techniques.

I pursued my Master's in Computer and Embedded Systems Engineering at TU Delft, where I worked with Dr. Guohao Lan and Dr. Xucong Zhang on immersive emotion recognition systems based on eye-tracking technology.

Personal Email  /  Work Email  /  CV  /  Scholar  /  Github

profile photo

Research

My interests drive by a passion for integrating AI to all conditions, such as embedded systems, wearable devices, etc. to create impactful, human-centered innovations. Some papers are highlighted.

Through the Eyes of Emotion: A Multi-faceted Eye Tracking Dataset for Emotion Recognition in Virtual Reality
Tongyun Yang†, Bishwas Regmi†, Lingyu Du, Andreas Bulling, Xucong Zhang, Guohao Lan
IMWUT/Ubicomp, 2025
paper / code / bibTex

An eye-tracking dataset in VR, combining high-frame-rate periocular videos and high-frequency gaze data to enable accurate, multimodal emotion recognition.

Reverse Imaging: Any-Sequence Generalization for Cardiac MRI Segmentation
Yidong Zhao, Yi Zhang, Tongyun Yang, Maša Božić-Iven, Ayda Arami, Yuchi Han, Orlando Simonetti, Hui Xue, Petter Kellman, Sebastian Weingärtner, Qian Tao
MICCAI & IEEE Transactions on Medical Imaging, 2025
paper / code / bibTex

A physics-driven framework that estimates tissue properties (M0, T1, T2) from annotated cardiac MRI images using diffusion models, enabling physics-based synthesis of diverse unseen sequences for zero-shot generalization of segmentation models across different MRI contrasts.

Pruning nnU-Net with Minimal Performance Loss
Tongyun Yang, Yidong Zhao, Qian Tao
MIDL, 2025
paper / code / bibTex

Trained nnU-Net models contain substantial weight redundancy, with over 80% of weights removable through simple magnitude-based pruning while maintaining same performance across multiple medical segmentation tasks.

Miscellanea

Teaching

TU Delft, ET 4310 Supercomputing for Big Data (2024/25 Q1) / TA
TU Delft, CESE 4030 Embedded Systems Lab (2023/24 Q3) / TA
TU Delft, CESE 4000 Software Fundamentals (2023/24 Q1) / TA
TU Delft, CESE 4010 Advanced Computing Systems (2023/24 Q1) / TA
TU Delft, Graduate Student Mentor (2023/24)

Website adopted from Jon Barron