Postdoctoral Researcher · Halıcıoğlu Data Science Institute, UC San Diego

Resource-efficient agents for science and engineering

AI agents can already generate work at scale. Turning that into real acceleration in science and engineering means accounting for what each step actually costs. I build the agents, the evaluations, and the benchmarks around that constraint. That instinct comes from particle physics, where every measurement is genuinely expensive and progress depends as much on engineering as on science.

Yue Ma

What I work on

Research →
  • 01Cost-aware agentic systemsClosed-loop systems that propose, implement, and evaluate experiments under real compute budgets.
  • 02Scientific agent evaluationRepresenting decision trajectories and action spaces, so agents can be judged on outcomes and on what they spent.
  • 03ML for scientific discoveryNeural processes and multi-fidelity surrogates for expensive simulation.
  • 04Statistical inference at scaleBayesian and likelihood methods for rare-event searches, and the data systems behind them.

What I build

Projects →
  • 01 · Agentic system

    SIDERIUS

    A closed-loop agentic system for autonomous research under real compute budgets.

    • Sole architect
    • Feb 2026 –
  • 02 · Methodology

    SciTra

    How an agent’s decision trajectory and action space are represented, so that runs are comparable across agents, tasks, and fields. 50+ scientists, 20+ institutions, with BenchFlow.

  • 03 · Domain benchmark

    FrontierPhysics

    One field, concretely: how AI agents carry out frontier physics research iteratively. With the BenchFlow team.

Selected publications

All publications →
  • SIDERIUSIn preparation Resource-aware multi-agent systems for autonomous machine-learning research. Project ↗
  • FrontierPhysicsIn preparation A benchmark for how AI agents carry out frontier physics research. benchflow.ai ↗
  • 2026 Resilient Load Forecasting under Climate Change: Adaptive Conditional Neural Processes Chenxi Hu, Yue Ma, Yifan Wu, Yunhe Hou · Preprint arXiv ↗
  • 2023 First Dark Matter Search with Nuclear Recoils from XENONnT XENON Collaboration · Phys. Rev. Lett. 131, 041003 Core contributor · event selection, efficiency modelling, background modelling
  • 2022 Search for New Physics in Electronic Recoil Data from XENONnT XENON Collaboration · Phys. Rev. Lett. 129, 161805 Core contributor · detector response calibration, data processing algorithms
  • 2026-09-02 SIDERIUS: Resource-Aware Multi-Agent System for Automated Machine LearningUpcoming Fast Machine Learning for Science Conference 2026, UC San Diego · Invited talk Programme ↗
  • 2026-03-16 TIDMAD: A Sandbox for Physics-Informed LLM Agents in Dark Matter Searches APS Global Physics Summit 2026 · Invited talk Slides ↗

Latest writing

All posts →