ARA Lab
Jiachen (Amber) Liu

Founder, ARA Lab

Jiachen “Amber” Liu

AI for Science · ML Systems

I’m building recursive self-improvement infrastructure: systems that get stronger the more they’re used.

Schedule a meeting ↗

About

Before ARA Lab, I got my PhD in Computer Science from the University of Michigan, advised by Prof. Mosharaf Chowdhury. My research focused on building efficient systems that let researchers push the boundary of machine learning, including LLM serving systems, LLM pre-training systems (Meta Llama & Avocado), and agentic RL.

News

Talks, workshops, and press.

  1. Validating AI-Driven Science convening by Schmidt Sciences in LondonSchmidt Sciences

    ConveningSchmidt Sciences · London

    Validating AI-Driven Science

    An invited convening with researchers from DeepMind, Anthropic, Hugging Face, eLife, and OpenAlex on how to check, reproduce, and build on science that AI proposes faster than we can test.

  2. Science x AI Summit 2026 at CaltechSAIR Foundation

    SpeakerSAIR Foundation · Caltech

    Science x AI Summit 2026

    Joining the Caltech lineup, with Barry Barish, Terence Tao, and Anima Anandkumar, on what AI actually changes about how science gets done.

  3. AI-Native Academia workshop at NeurIPS 2026NeurIPS

    WorkshopNeurIPS 2026 · Atlanta

    AI-Native Academia

    The workshop where the program chairs of ICLR, ICML, and CVPR redesign authorship, peer review, and conference governance under AI. Sponsored by ARA Lab.

  4. Talk on Agent-Native Research Artifacts at LossfunkLossfunk

    TalkLossfunk

    Agent-Native Research Artifacts

    A full walk through the standard for the Lossfunk research community: the two taxes, the four layers, and what an agent does with them.

  5. QbitAI coverage of Agent-Native Research Artifacts量子位 QbitAI

    Press量子位 QbitAI

    PDF当死，ARA该立！论文是时候Agent原生了

    One of China’s largest AI outlets on the storytelling and engineering taxes, and on what replaces the PDF.

  6. IEEE Spectrum interview on the research paper formatIEEE Spectrum

    PressIEEE Spectrum

    Should Researchers Write Papers for AI Instead of People?

    David Berreby interviews me on replacing the PDF with a format written for the machine that reads it first.

  7. Synced coverage of The Last Human-Written Paper机器之心 Synced

    Press机器之心 Synced

    「这可能是人类写的最后一篇论文」

    The first outlet to cover the paper: 37 researchers across Stanford, Michigan, and CMU rewriting the PDF into a research package an AI can execute directly.

Publications

Selected papers.

AI for Science

  1. arXiv 2026

    The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement

    Yi Duan, ..., Jiachen Liu, ..., Fan Wu (36 authors, Shanghai Jiao Tong University-led collaboration)

    Paper

  2. arXiv 2026

    Beyond Natural Language: An Agent-Native Language for Autonomous Science

    Yifeng He, Jiachen Liu

    Paper

  3. NeurIPS 2026 AI for Science Workshop

    The Greatness of Science Cannot Be Planned: Agentic Auto-Research is Fuzz Testing

    Yifeng He, Jicheng Wang, Yinzhe Zhao, Chengyang Shi, Jiachen Liu, Hao Chen

    Paper

  4. NeurIPS 2026ICML & CAIS Workshop Spotlight

    The Last Human-Written Paper: Agent-Native Research Artifacts

    Jiachen Liu, Jiaxin Pei, Jintao Huang, Chenglei Si, Ao Qu, Xiangru Tang, Runyu Lu, Lichang Chen, Xiaoyan Bai, Haizhong Zheng, Carl Chen, Zhiyang Chen, Haojie Ye, Yujuan Fu, Zexue He, Zijian Jin, Zhenyu Zhang, Shangquan Sun, Maestro Harmon, Dianzhuo Wang, Jianqiao Zeng, Jiachen Sun, Mingyuan Wu, Baoyu Zhou, Chenyu You, Shijian Lu, Yiming Qiu, Fan Lai, Yuan Yuan, Yao Li, Junyuan Hong, Ruihao Zhu, Beidi Chen, Alex Pentland, Ang Chen, Mosharaf Chowdhury, Zechen Zhang

    PaperBlog

  5. arXiv 2026

    Sci-Reasoning: A Dataset Decoding AI Innovation Patterns

    Jiachen Liu, Maestro Harmon, Zechen Zhang

    PaperDataset

  6. ICLR 2026

    EXP-Bench: Can AI Conduct AI Research Experiments?

    Patrick Tser Jern Kon*, Jiachen Liu*, Xinyi Zhu, Qiuyi Ding, Jingjia Peng, Jiarong Xing, Yibo Huang, Yiming Qiu, Jayanth Srinivasa, Myungjin Lee, Mosharaf Chowdhury, Matei Zaharia, Ang Chen (* Equal contribution)

    PaperBlogDataset

  7. arXiv 2025

    Curie: Toward Rigorous and Automated Scientific Experimentation with AI Agents

    Patrick Tser Jern Kon*, Jiachen Liu*, Qiuyi Ding, Yiming Qiu, Zhenning Yang, Yibo Huang, Jayanth Srinivasa, Myungjin Lee, Mosharaf Chowdhury, Ang Chen (* Equal contribution)

    GitHubPaperBlog & Demo

Machine Learning Systems

  1. PhD Dissertation 2025

    User-Centric Machine Learning Systems

    Jiachen Liu

    PaperBlog

  2. NeurIPS 2025Spotlight

    The ML.ENERGY Benchmark: Toward Automated Inference Energy Measurement and Optimization

    Jae-Won Chung, Jiachen Liu, Jeff Ma, Ruofan Wu, Oh Jun Kweon, Yuxuan Xia, Zhiyu Wu, Mosharaf Chowdhury

    DemoPaper

  3. MLSys 2025

    Venn: Resource Management for Collaborative Learning Jobs

    Jiachen Liu, Fan Lai, Ding Ding, Yiwen Zhang, Mosharaf Chowdhury

    Paper

  4. arXiv 2024

    Andes: Defining and Enhancing Quality-of-Experience in LLM-Based Text Streaming Services

    Jiachen Liu, Zhiyu Wu, Jae-Won Chung, Fan Lai, Myungjin Lee, Mosharaf Chowdhury

    BlogPaper

  5. TMLR 2024

    Efficient Large Language Models: A Survey

    Zhongwei Wan, Xin Wang, Che Liu, Samiul Alam, Yu Zheng, Jiachen Liu, Zhongnan Qu, Shen Yan, Yi Zhu, Quanlu Zhang, Mosharaf Chowdhury, Mi Zhang

    PaperGitHub

  6. MLSys 2021

    Fluid: Resource-Aware Hyperparameter Tuning Engine

    Peifeng Yu*, Jiachen Liu*, Mosharaf Chowdhury (* Equal contribution)

    GitHubPaperVideo

Others

  1. Nature 2026

    Humanity’s Last Exam

    Long Phan, ..., Jiachen Liu, ..., Dan Hendrycks (Scale AI & CAIS collaborative benchmark)

    PaperAnnouncement

  2. MLSys 2024

    FedTrans: Efficient Federated Learning via Multi-Model Transformation

    Yuxuan Zhu, Jiachen Liu, Mosharaf Chowdhury, Fan Lai

    Paper

  3. SoCC 2023

    Auxo: Efficient Federated Learning via Scalable Cohort Identification

    Jiachen Liu, Fan Lai, Yinwei Dai, Aditya Akella, Harsha Madhyastha, Mosharaf Chowdhury

    GitHubPaper

  4. ICML 2022

    FedScale: Benchmarking Model and System Performance of Federated Learning at Scale

    Fan Lai, Yinwei Dai, Sanjay S. Singapuram, Jiachen Liu, Xiangfeng Zhu, Harsha V. Madhyastha, Mosharaf Chowdhury

    GitHubPaper

  5. arXiv 2025

    Evaluation Framework for AI Systems in “the Wild”

    Sarah Jabbour, Trenton Chang, Anindya Das Antar, Joseph Peper, Insu Jang, Jiachen Liu, Jae-Won Chung, Shiqi He, Michael Wellman, Bryan Goodman, Elizabeth Bondi-Kelly, Kevin Samy, Rada Mihalcea, Mosharaf Chowdhury, David Jurgens, Lu Wang

    Paper

  6. NeurIPS 2024

    IaC-Eval: A code generation benchmark for Infrastructure-as-Code programs

    Patrick Tser Jern Kon, Jiachen Liu, Yiming Qiu, Weijun Fan, Ting He, Lei Lin, Haoran Zhang, Owen M. Park, George Sajan Elengikal, Yuxin Kang, Ang Chen, Mosharaf Chowdhury, Myungjin Lee, Xinyu Wang

    GitHubPaper

Awards

Recognition.

  • Machine Learning and Systems Rising Stars (Program of 2023)
  • Shanghai (China) Outstanding Graduate (Top 5%), 2020

Experience

Where I’ve worked.

  1. ARA LabPalo Alto, CACo-FounderBacked by DCVC.
  2. Meta Superintelligence LabMenlo Park, CAResearch ScientistLLM pre- and post-training systems.
  3. Meta (AI & Systems Co-Design)Menlo Park, CAPart-time Research ScientistCross-regional Llama training systems.
  4. University of MichiganAnn Arbor, MITeaching AssistantSystems for Generative AI (EECS 598)
  5. AppleNew York, NYPhD InternPrivate machine learning framework.
  6. MIT CSAIL Database GroupCambridge, MAResearch InternHigh-dimensional vector database. Advised by Prof. Sam Madden
  7. University of MichiganAnn Arbor, MITeaching AssistantDatabase Management (EECS 484)

Projects

Open-source systems I helped build.

Orchestra

Orchestra

Vibe Research Platform

An AI-native platform for Vibe Research, built to take scientific work from idea to publication.

Curie2024

Curie

Experiment Agent for Scientific Research

The first AI-agent framework designed for automated and rigorous scientific experimentation.

ML.Energy2023

ML.Energy Leaderboard

Energy Benchmark for LLMs

Gives people a sense of how much energy LLMs consume, measured rather than guessed.

FedScale2022

FedScale

Federated Learning Engine

A scalable and extensible open-source federated learning (FL) engine and benchmark.

Community Service

Service.

  1. UMich CSE Graduate Students OrganizationAnn Arbor, MIDEI Chair
  2. SJTU Joint Institute Student UnionShanghai, ChinaVice President

Education

Degrees.

  1. University of MichiganAnn Arbor, MIMaster & Ph.D. in Computer Science
  2. University of MichiganAnn Arbor, MIB.S.E. in Data Science
  3. Shanghai Jiao Tong UniversityShanghai, ChinaB.E. in Electrical and Computer Engineering

Contact

Jiachen Liu

Personal homepage ↗Back to the team