I am a Ph.D. candidate at Tsinghua University, advised by Prof. Gao Huang at LeapLab. I received my B.Eng. from Tsinghua University in 2023. I am currently a Top Seed Intern at ByteDance Seed.
My research centers on efficient and scalable architecture design, with a focus on test-time training, sequence modeling, context scaling, and continual learning. My work ViT3: Unlocking Test-Time Training in Vision was selected as a CVPR 2026 Best Paper Finalist (15 / 16,092 submissions) and an Oral presentation.
News
| Oct 2026 | 🏆 Honored to receive the National Scholarship for the second consecutive year, in 2026! |
|---|---|
| Jun 2026 | 🏆 Our paper ViT3: Unlocking Test-Time Training in Vision is a CVPR 2026 Best Paper Finalist (15 / 16,092 submissions) and an Oral presentation! |
| May 2026 | 🎉 Two papers were accepted at ICML 2026! |
| Jan 2026 | ✨ I joined ByteDance Seed as a Top Seed Intern, working on LLM architecture research with a focus on context scaling and continual learning. |
| Oct 2025 | 🏆 Honored to receive the National Scholarship in 2025! |
Selected Publications
- CVPR
ViT3: Unlocking Test-Time Training in VisionIn IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026OralBest Paper Finalist15/16092 - AAAI
Vision Transformers are Circulant Attention LearnersIn AAAI Conference on Artificial Intelligence (AAAI), 2026 - NeurIPS
Demystify Mamba in Vision: A Linear Attention PerspectiveIn Advances in Neural Information Processing Systems (NeurIPS), 2024 - NeurIPS
Bridging the Divide: Reconsidering Softmax and Linear AttentionIn Advances in Neural Information Processing Systems (NeurIPS), 2024 - ICCV
FLatten Transformer: Vision Transformer using Focused Linear AttentionIn IEEE/CVF International Conference on Computer Vision (ICCV), 2023
Internship Experience
-
Top Seed Intern
2026.01 – PresentByteDance Seed - Research on large language model architectures, focusing on context scaling and continual learning.
-
Research Intern
2024.05 – 2025.12Alibaba Future Life Lab - Research on advanced efficient sequence modeling mechanisms.
Honors and Awards
| 2026 |
|
|---|---|
| 2025 |
|
| 2023 |
|
| 2022 |
|
| 2021 |
|
| 2020 |
|
| 2019 |
|