Dual Domain-attribute Learning Framework with Asynchronous Adapters for Continual Test-time Adaptation
Yuntong Tian, Kang Li, Tianyang He, Liang Wan, Pheng-Ann Heng, Wei Feng.
Computer Science Undergraduate | Tianjin University
Tianyang He ไฝๅคฉ้ณ
Hi, I'm Tianyang He (Steven). I work across machine learning, computer vision, and systems, with a focus on robustness under distribution shift and efficient adaptation.
Machine Learning
Computer Vision
Robustness
Continual Adaptation
Compilers
About
What I focus on day-to-day.
Core Focus
I explore robust and adaptive vision models, especially continual test-time adaptation, domain shift robustness, and parameter-efficient tuning.
Working Style
I enjoy combining theory, large-scale experiments, and systems implementation, from adapter-based vision models to C/LLVM compilers and OS kernels.
Publications
Selected work
Research Interests
Current directions
Robust & Adaptive Vision
- Continual test-time adaptation (CTTA)
- Domain generalization & distribution shift
- Adapter-based and LoRA-style tuning
Foundation Models & Feedback
- MLLM-assisted self-correction loops
- Reliable feedback and verification
Systems & Programming Languages
- Compilers: front-end, IR, optimization
- OS-level support for AI workloads
- GPU cluster performance engineering