About Me Flower

Hi, I'm Xiyu Liang.

I obtained my bachelor's degree from University of Electronic Science and Technology of China (UESTC) . Recently, I am a research intern at Georgia Tech , under the guidance of Prof. Yingyan (Celine) Lin . Previously, I was a research assistant in the Spatio-Temporal Big Data and Intelligence Lab, where I worked on spatio-temporal data mining under the supervision of Prof. Shuo Shang and Prof. Lisi Chen. I am also a member of ColAI (Now DreamSoul-AI , an AI startup), where I work closely with Dr. Enmao Diao.

My research interests primarily lie in machine learning and natural language processing. I am particularly passionate about building large-scale machine learning models and systems that achieve lower cost, higher speed, and better performance. Additionally, I am also exploring trustworthy AI, to make AI systems more reliable and accessible for humans.

🎓 Education
  • May 2023 - Aug. 2023
    National University of Singapore (NUS)
    National University of Singapore (NUS)
    Visiting Student at School of Computing
  • Sep. 2021 - Jun. 2025
    University of Electronic Science and Technology of China (UESTC)
    University of Electronic Science and Technology of China (UESTC)
    B.Eng in Software Engineering
📍 Experience
  • Oct. 2025 - Present
    Georgia Institute of Technology, EIC Lab
    Georgia Institute of Technology, EIC Lab Georgia Institute of Technology, EIC Lab 2
    Research Intern
    Topic: Efficient Diffusion LLMs
  • May 2024 - Sept. 2025
    Duke University, ColAI Group
    Duke University, ColAI Group Duke University, ColAI Group 2
    Research Assistant
    Topic: KV Cache Compression
  • Oct. 2023 - May 2025
    UESTC, Spatio-Temporal Big Data and Intelligence Lab
    UESTC, Spatio-Temporal Big Data and Intelligence Lab
    Research Assistant
    Topic: Efficient Online Trajectory Clustering
🏆 Honors & Awards
  • UESTC First-Class Academic Scholarship
    2024
  • National Second Prize (Top 0.3%, 20/5801), awarded 2000 CNY
    2024
    Chinese Software Cup
  • Gold Award (Top prize)
    2023
    9th Sichuan International "Internet+" College Student Innovation and Entrepreneurship Competition
  • Regional Third Prize (Southwest China)
    2023
    14th China College Students' Service Outsourcing Innovation and Entrepreneurship Competition
Publications (view all )
OBCache: Optimal Brain KV Cache Pruning for Efficient Long-Context LLM Inference
OBCache: Optimal Brain KV Cache Pruning for Efficient Long-Context LLM Inference

Yuzhe Gu, Xiyu Liang, Jiaojiao Zhao, Enmao Diao

Under review. 2026

Optimal Brain Cache (OBCache), a principled framework that formulates KV cache eviction as a layer-wise structured pruning problem. OBCache quantifies token saliency by measuring the perturbation in attention outputs induced by pruning tokens, with closed-form scores derived for isolated keys, isolated values, and joint key-value pairs.

OBCache: Optimal Brain KV Cache Pruning for Efficient Long-Context LLM Inference

Yuzhe Gu, Xiyu Liang, Jiaojiao Zhao, Enmao Diao

Under review. 2026

Optimal Brain Cache (OBCache), a principled framework that formulates KV cache eviction as a layer-wise structured pruning problem. OBCache quantifies token saliency by measuring the perturbation in attention outputs induced by pruning tokens, with closed-form scores derived for isolated keys, isolated values, and joint key-value pairs.

All publications