Be open to changes, and learn from the best.

• 4th year Ph.D. candidate at the Institute for AI Industry Research (AIR), Tsinghua University.
  • Advised by Prof. Yunxin Liu（刘云新）, working closely with Prof. Ting Cao（曹婷）.
• Received my B.E. degree from the Department of Electronic Engineering (EE), Tsinghua Univeristy in 2022.
• Interested in on-device and physical AI. Previous works focused on system efficiency:
  • OxyGen: unified KV cache management for multi-task VLA inference. [paper] [code]
  • Vec-LUT (MobiSys 2026): LUT-based mpGeMM kernel for ultra-low-bit LLM inference. [paper] [code] [slides] [model]
  • FlexNN (MobiCom 2024): memory-adaptive DNN inference on edge devices. [paper] [code] [slides]
  • Personal LLM Agent (Survey, "Efficiency" Section Lead). [paper] [repo]
• Currently focused on algorithm-system co-design toward self-evolving embodied agents.

Looking for opportunities in Embodied AI & AI Infra. Feel free to email me: lixiangy22@mails.tsinghua.edu.cn.

• [Jun. 2026] Vec-LUT received Best Paper Award Runner-Up at MobiSys 2026.
• [May. 2026] OxyGen extends PyTorch backend (previously JAX-only) for on-board deployment (e.g., Jetson AGX Thor).

• Google Scholar
• Homepage
• 小红书：Xyu（3612640537）