Hong T. Nguyen
USC. Los Angeles. 3237879669. Live a life you will never regret.
RTH 318, USC
Los Angeles, CA 90089
I am a Ph.D. student in Electrical and Computer Engineering at the University of Southern California (USC), expected to graduate in December 2027. I am a Research Assistant in the Signal Analysis and Interpretation Laboratory (SAIL), advised by Prof. Shrikanth Narayanan. Previously, I received my Bachelor’s degree from Hanoi University of Science and Technology (HUST), supervised by Prof. Chuyen Nguyen Thanh.
My research focuses on human-centered video understanding and temporal dynamics, in the context of medical imaging, clinical trials, and seamless dyadic interaction. I develop explainable multimodal models of human behavior across video and speech, including video world models and diffusion models for real-time MRI of the vocal tract (Arti-JEPA, Speech2rtMRI), motion-centric video encoders (MOOSE), and severity representation learning for medical images (ConPro). I work with multidisciplinary teams spanning law enforcement, healthcare, linguistics, and psychology. Feel free to reach out for research collaboration at hongn@usc.edu.
Beyond technical expertise, I am also boardgame lover and hiker. Yet, I have only visited 4/63 National Parks in the US.
news
| Sep 09, 2026 | New preprint: Arti-JEPA adapts a video world model to real-time MRI of the vocal tract for speech-production analysis. Read it on arXiv |
|---|---|
| May 03, 2026 | Our paper on interpretable modeling of articulatory temporal dynamics from real-time MRI for phoneme recognition appears at ICASSP 2026. |
| Jun 01, 2025 | Preprint of MOOSE, a motion-centric video encoder that pays attention to temporal dynamics via optical flow, is out on arXiv. |
| Apr 06, 2025 | My paper Speech2rtMRI, the first speech-guided diffusion model for real-time MRI video of the vocal tract, appears at ICASSP 2025 |
| Oct 24, 2024 | This website is on air... |
selected publications
- arXivMOOSE: Pay Attention to Temporal Dynamics for Video Understanding via Optical FlowsarXiv preprint arXiv:2506.01119, 2025