I am a robotics engineer at Mujin@Tokyo. I have been working broadly on computer vision, artificial intelligence, and robotics. I am now also looking for PhD opportunities starting in 2027 or 2028 
During my undergrad, I was a research student at Pair Lab@Georgia Tech, advised by Prof. Animesh Garg. I am very fortunate to have worked with him and Dr. Wei Yu.
On the industry side, I have interned at Amazon@Seattle and SMBC@NYC, and worked at Tencent@Beijing as a backend infrastructure engineer.
I have lived in Nanjing, Toronto, Vancouver, New York, Beijing, and now Tokyo. In my free time, I enjoy landscape photography and learning languages. I always love making new friends and finding ways to work together — feel free to reach out~ 
Warning
Problem: The current name of your GitHub Pages repository ("Solution: Please consider renaming the repository to "
http://".
However, if the current repository name is intended, you can ignore this message by removing "{% include widgets/debug_repo_name.html %}" in index.html.
Action required
Problem: The current root path of this site is "baseurl ("_config.yml.
Solution: Please set the
baseurl in _config.yml to "Education
-
Columbia UniversityM.Sc. in Computer Science2022 - 2024 -
University of TorontoB.Sc. in Computer Science2018 - 2022
Honors & Awards
-
Norman Stuart Robertson Scholarship (Top 3 in math major)
-
C.L. Burton Scholarship for Mathematics and Physical Sciences (2 times)
-
University of Toronto Research Excellence Award (UTEA)
-
College Silver Medal
Publications
Conferences
EgoSim: Egocentric Exploration in Virtual Worlds with Multi-modal Conditioning
Wei Yu, Songheng Yin, Steve Easterbrook, Animesh Garg
International Conference on Learning Representations (ICLR) 2025
EgoSim generates egocentric exploration videos of virtual worlds conditioned on multi-modal inputs.
EgoSim: Egocentric Exploration in Virtual Worlds with Multi-modal Conditioning
Wei Yu, Songheng Yin, Steve Easterbrook, Animesh Garg
International Conference on Learning Representations (ICLR) 2025
EgoSim generates egocentric exploration videos of virtual worlds conditioned on multi-modal inputs.
Modular Action Concept Grounding in Semantic Video Prediction
Wei Yu, Wenxin Chen, Songheng Yin, Steve Easterbrook, Animesh Garg
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2022
We propose modular action concept grounding (MAC), which represents actions as compositions of grounded semantic concepts to improve semantic video prediction.
Modular Action Concept Grounding in Semantic Video Prediction
Wei Yu, Wenxin Chen, Songheng Yin, Steve Easterbrook, Animesh Garg
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2022
We propose modular action concept grounding (MAC), which represents actions as compositions of grounded semantic concepts to improve semantic video prediction.
Preprint / Under Review
When ODE Fidelity Misleads Diffusion Timestep Search
Songheng Yin*, Lanfeng Wang*, Wei Yu, Feng Yuan, Souleyman Boudouh, Qi Luo, Youming Tao (* equal contribution)
under review
When ODE Fidelity Misleads Diffusion Timestep Search
Songheng Yin*, Lanfeng Wang*, Wei Yu, Feng Yuan, Souleyman Boudouh, Qi Luo, Youming Tao (* equal contribution)
under review
Peers as Teachers: Filling Supervision Gaps in Single-Robot Occupancy Forecasting
Lanfeng Wang*, Songheng Yin*, Yuan Feng, Zhao-Han Peng, Rui-Qi Dong (* equal contribution)
under review
Peers as Teachers: Filling Supervision Gaps in Single-Robot Occupancy Forecasting
Lanfeng Wang*, Songheng Yin*, Yuan Feng, Zhao-Han Peng, Rui-Qi Dong (* equal contribution)
under review
AAI-BWM: Action-Conditioned Acoustic Innovation for Audio-Visual World Modeling
Lanfeng Wang, Zhao-Han Peng, Jiayang Mao, Li Wang, Songheng Yin, Zhengxian Chen
under review
AAI-BWM: Action-Conditioned Acoustic Innovation for Audio-Visual World Modeling
Lanfeng Wang, Zhao-Han Peng, Jiayang Mao, Li Wang, Songheng Yin, Zhengxian Chen
under review
QUOTA: Quantization via Output-channel Targeted Allocation
Souleyman Boudouh, Simla Burcu Harma, Alireza Khodamoradi, Abdulrahman Mahmoud, Songheng Yin, Kristof Denolf, Babak Falsafi
under review
Learned, per-channel mixed-precision budget can largely amortize the collapse across diverse backbones and scales.
QUOTA: Quantization via Output-channel Targeted Allocation
Souleyman Boudouh, Simla Burcu Harma, Alireza Khodamoradi, Abdulrahman Mahmoud, Songheng Yin, Kristof Denolf, Babak Falsafi
under review
Learned, per-channel mixed-precision budget can largely amortize the collapse across diverse backbones and scales.
MosaicMem: Hybrid Spatial Memory for Controllable Video World Models
Wei Yu, Runjia Qian, Yumeng Li, Liquan Wang, Songheng Yin, Sri Siddarth Chakaravarthy P, Dennis Anthony, Yang Ye, Yidi Li, Weiwei Wan, Animesh Garg
arXiv preprint 2026
MosaicMem introduces a hybrid spatial memory that enables controllable video world models.
MosaicMem: Hybrid Spatial Memory for Controllable Video World Models
Wei Yu, Runjia Qian, Yumeng Li, Liquan Wang, Songheng Yin, Sri Siddarth Chakaravarthy P, Dennis Anthony, Yang Ye, Yidi Li, Weiwei Wan, Animesh Garg
arXiv preprint 2026
MosaicMem introduces a hybrid spatial memory that enables controllable video world models.