Peiyuan Liao


Researcher, Programmer, Ponderer

When in doubt, Unsafe.Coerce

I help organize an academic conference/retreat in Beijing.

I also write code for a Pittsburgh-Chicago-based startup on real-time capture solutions for virtual personalities and youtubers/livers/“UP主” (uploaders).

While at school, I focused/am focusing on topics that live in the intersection of machine learning, systems engineering and programming language theory. 2 I've had the honor of working with Han Zhao and Keyulu Xu on graph neural networks in the past.

2. For a complete list of published works and preprints, please refer to papers.bib.

Research Interests

  • Graph Neural Networks: performance, application and interpretability
  • Representation Learning
  • Machine Learning/Deep Learning Systems: Training and Inference DSLs, Compilers and Optimizers
  • Federated Learning
  • Distributed Systems
  • Object Detection and Segmentation under performance-sensitive settings (low-power, low-memory, etc.)
  • Disentangled Machine Learning/Meta Learning/Interpretable Machine Learning
  • Formal Verification of Machine Learning Algorithms
  • Automated Abstraction and Reasoning of Programs
  • Currently

    I'm mainly participating in Recurse Center's Summer I batch (2020), where I will work several open source projects, including mmdetection, TVM, Taichi and PyTorch Geometric.

    I am also working at Cambricon Technologies as an intern, where I seek to enhance core features of their MLU enviornment and the BANG programming language.

    Coursework (Expected)

  • 15-295 Competitive Programming
  • 10-315 Introduction to Machine Learning
  • 15-459 Quantum Computation
  • 15-418 Parallel Architecture and Programming
  • 15-819 Foundation of Quantitative Program Analysis (graduate)
  • Portfolio

    I'm involved in several online, computer science-related communities, including Kaggle and Codewars (real-time ranking below). My resume can be found here.

    Codewars Profile

    graph neural networks, lossy compression


    graph neural networks, lossy compression
    functional programming, deep learning DSL, data science


    functional programming, deep learning DSL, data science
    classes, majors, courses


    classes, majors, courses
    LunTan, Recurse Center, Cambricon Technologies


    LunTan, Recurse Center, Cambricon Technologies
    programming languages, natural languages, technologies


    programming languages, natural languages, technologies
    prizes and recognitions


    prizes and recognitions


    Personal Interests

  • (hate to say it, but) towards General Artificial Intelligence
  • Automated Theorem Proving
  • Coldplay
  • Analytic Philosophy & its application in computer science
  • Automata Theory
  • Gen Hoshino
  • Sakanaction
  • French/Japanese
  • Film Photography

  • Prague
  • Virtual YouTubers *
  • Squash
  • *. 小孩子干什么不好非得看管人


    Regarding my researches, academic collaboration, Kaggle team-up, or questions related to TVM/zeta/ml-arsenal:

    peiyuanl [at] andrew [dot] cmu [dot] edu

    For LunTan-related subjects, such as application, collaboration, sponsorship, etc:

    peiyuanl [at] luntan [dot] io

    Finally, anything else, like just making friends or talk about programming languages, machine learning (for fun), or what's it's like to be at recurse center:

    alexander [underscore] liao [at] outlook [dot] com

    Most of the code I've written can be found under my GitHub account: @liaopeiyuan.

    Institutional Address

    Peiyuan Liao
    5032 Forbes Avenue
    SMC 4401
    Pittsburgh, PA 15289

    ORCID iD icon

    tr5 Homepage

    Want to become a better programmer? Join the Recurse Center!