avatar

Xuting Liu

Ph.D. Student
Computer and Information Science
University of Pennsylvania
xutingl [at] seas.upenn.edu


Hi I’m Xuting Liu

I’m a Ph.D. student at UPenn’s Distributed Systems Lab advised by Vincent Liu.

Before joining UPenn, I got my bachelor’s degree at UC Berkeley with a double major in Computer Science and Operations Research, and interned at the Sky Lab advised by Joseph Gonzalez and Paras Jain.

Check out my CV here!

Always open to research collaboration, mentoring, and exciting discussions! Feel free to reach out via email.

Research Interests

Machine learning systems for: Early-Exit / Speculative Decoding / Looped Transformers;

Distributed systems: PD disaggrgation, collective communication.

Preprints

  1. Dynamic Rebatching for Efficient Early-Exit Inference with DREX
    Xuting Liu, Daniel Alexander, Siva Kesava Reddy Kakarla, Behnaz Arzani, Vincent Liu
    arXiv

Publications

  1. 𝝀-trim: Optimizing Function Initialization in Serverless Applications With Cost-driven Debloating
    Xuting Liu*, Spyros Pavlatos*, Yuhao Liu, Vincent Liu (* Equal contribution)
    International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS), 2025.
    Paper Code

  2. Rethinking Machine Learning Collective Communication as a Multi-Commodity Flow Problem (TECCL)
    Xuting Liu, Behnaz Arzani, Siva Kesava Reddy Kakarla, Liangyu Zhao, Vincent Liu, Miguel Castro, Srikanth Kandula, Luke Marshall
    ACM Special Interest Group on Data Communication (SIGCOMM), 2025.
    Paper Code

Work Experience

  1. Google
    Systems Research Group
    11/2025 - Present
    TPU collective communication with ICI and Pallas.
  2. Amazon
    Deep Science for Systems and Services
    08/2025 - 10/2025
    Prefill-Decode disaggregated LLM inference.
  3. Microsoft Research Asia, Beijing
    07/2024 - 10/2024
    Topologiy optimization for distributed ML.
  4. Microsoft Research, Redmond
    12/2023 - 12/2024
    GPU collective communication.

Talks

  1. TECCL: Rethinking Machine Learning Collective Communication as a Multi-Commodity Flow Problem
    ByteDance, September 2024. Network and Mobile Systems (NMS) group, MIT CSAIL, March 2025.


Powered by Jekyll and Minimal Light theme.