HTTP/2 200 server: GitHub.com content-type: text/html; charset=utf-8 last-modified: Tue, 16 Dec 2025 06:23:55 GMT access-control-allow-origin: * strict-transport-security: max-age=31556952 etag: W/"6940fafb-5882" expires: Sun, 28 Dec 2025 21:28:19 GMT cache-control: max-age=600 content-encoding: gzip x-proxy-cache: MISS x-github-request-id: CE9D:2D8B9D:7FEAE7:8F91C0:69519E9B accept-ranges: bytes age: 0 date: Sun, 28 Dec 2025 21:18:19 GMT via: 1.1 varnish x-served-by: cache-bom-vanm7210096-BOM x-cache: MISS x-cache-hits: 0 x-timer: S1766956699.492583,VS0,VE206 vary: Accept-Encoding x-fastly-request-id: 793895ff2f2aef60e2d0ae35e6ca4268ed9c1ad5 content-length: 6077

• Email copied
• Linkedin
• Github
• Twitter

Jinghan Ke 

---

I am a CS Ph.D. student (25Fall) at UT Austin, co-advised by Prof. Qixing Huang and Prof. Atlas Wang.

Previously, I was a research assistant at UC Berkeley, working with Prof. Angjoo Kanazawa, Prof. Kurt Keutzer, and Prof. Masayoshi Tomizuka. Prior to that, I worked at NUS under Prof. Lin Shao. I received my bachelor's degree from the USTC.

Selected Publications | Google Scholar

---

Angles Don't Lie: Unlocking Training-Efficient RL Through the Model's Own Signals

Wang Qinsi, Jinghan Ke, Hancheng Ye, Yueqian Lin, Yuzhe Fu, Jianyi Zhang, Kurt Keutzer, Chenfeng Xu^✝, Yiran Chen^✝ (^✝ corresponding author.)

Neural Information Processing Systems (NeurIPS) 2025, Spotlight

Paper

Poster

Code

Dobi-SVD: Differentiable SVD for LLM Compression and Some New Perspectives

Wang Qinsi*, Jinghan Ke*, Masayoshi Tomizuka, Yiran Chen, Kurt Keutzer, Chenfeng Xu^✝ (* equal contribution. Order is decided by coin flip. ^✝ corresponding author.)

International Conference on Learning Representations (ICLR) 2025

Paper

Page

Code

MathNAS: If Blocks Have a Role in Mathematical Architecture Design

Wang Qinsi*, Jinghan Ke* (* equal contribution)

Neural Information Processing Systems (NeurIPS) 2023

PDF

Poster

Code

Diff-LfD: Contact-aware Model-based Learning from Visual Demonstration for Robotic Manipulation via Differentiable Physics-based Simulation and Rendering 

Xinghao Zhu, Jinghan Ke, Zhixuan Xu, Zhixin Sun, Bizhe Bai, Jun Lv, ..., Cewu Lu, Masayoshi Tomizuka, Lin Shao

Conference on Robot Learning (CoRL) 2023, Oral

PDF

Page

Cite

Citation copied to clipboard!

Video Editing | Youtube

---

When I'm not doing research, I like to collect rare experiences and skills. For example, I once spent eight days cycling solo around Hainan Island—about 1,000 km— and I can perform Chinese sword dance (intro).

If I have more free time, being “on the road” is my default: I wander through the corners of different cities and watch how people live; there are always unexpected people and moments around the next turn.

In truth, I mostly remain ignorant. Reading as I traveled a few years ago, after visiting thirty or forty cities, I realized I still don't understand my own country. Likewise—even if one day I visit a hundred countries, I still wouldn't claim to understand this planet; and in some farther future, even if I visit a hundred Earth-like planets, I still wouldn't claim to understand the universe.