PhD candidate / AI researcher
I am a PhD candidate in Computer Science and Artificial Intelligence at Oregon State University, advised by Liang Huang. My research uses sign-language translation and structured 3D human motion as demanding settings for multilingual generation, streaming inference, and evaluation. I have also worked on long-context multimodal models and text-to-SQL at Amazon and Genies.
Seeking full-time Research Scientist, Applied Scientist, Research Engineer, and ML Engineer roles starting in 2027.
01 / Selected work
Direct ASL, CSL, and DGS translation without a spoken-text pivot. The system uses 112K cross-lingual pairs and a shared motion representation, reducing mean motion error by 19% and running 2.3× faster than a three-stage cascade.
Multilingual generation · mBART · VQ-VAE · synthetic data · evaluation
Streaming sign-to-sign generation with prefix training and wait-k decoding. The approach cuts computation-aware latency by 38% while making the quality–latency trade-off explicit.
Streaming decoding · SMPL-X · wait-k · multi-path training
Variable-length 3D sign generation with geometry-aware objectives for articulated hands and consistent skeletons. The model improves back-translation BLEU-4 by 3.08 points on PHOENIX14T development data.
3D human motion · XLM-R · OpenPose · inverse kinematics
02 / Industry
2024–2025
Built an LLM-based user-insight system and explored GRPO and MCTS-guided reasoning for text-to-SQL.
2023
Researched generative models for long multimodal sequences and cross-modal dependencies.
2022
Developed multilingual representation-learning methods on XLM-R for cross-lingual retrieval.
2019–2020
Built generative sequence models for music style transfer and MIDI generation.
03 / Writing
Direct Translation between Sign Languages
Zetian Wu, Bowen Xie, Wuyang Meng, Milan Gautam, Stefan Lee, and Liang Huang
Geometry-Aware Losses for Structure-Preserving Text-to-Sign Language Generation
Zetian Wu, Tianshuo Zhou, Stefan Lee, and Liang Huang
Inducing Generalizable and Interpretable Lexica
Findings of EMNLP
MultiBench: Multiscale Benchmarks for Multimodal Representation Learning
NeurIPS Datasets and Benchmarks
Interactive Rare-Category-of-Interest Mining from Large Datasets
AAAI
Education. PhD, Oregon State University · MSE, Johns Hopkins University · BS, Zhejiang University.
Tools. Python, PyTorch, Hugging Face, SQL, C/C++, multi-GPU training, LLM post-training, multimodal learning, and 3D human motion.