Physicist & AI Researcher
Fan Xiaoyan 樊孝衍
Particle Physics · Machine Learning
B.Sc. Physics, University of Malaya M.Phil. AI, The Hong Kong University of Science and Technology (Guangzhou)
“Setbacks are not stumbling blocks but stepping stones.”
Physicist & AI Researcher
Particle Physics · Machine Learning
B.Sc. Physics, University of Malaya M.Phil. AI, The Hong Kong University of Science and Technology (Guangzhou)
“Setbacks are not stumbling blocks but stepping stones.”
Roughly 95% of the universe is dark matter and dark energy — the very scaffolding of everything that exists — and we still have no idea what these fundamental LEGO bricks actually look like. That mystery is what pulled me into physics, and it still does.
My curiosity about nature also shapes my interest in intelligence. I am drawn to how physics can guide reliable world models, and how agents can learn through interaction and improve their own capacity to explore. My long-term ambition is to study intelligence that poses its own questions, tests its ideas against the world, and discovers natural laws that remain unknown to us.
Alongside this scientific ambition, I hope AI can make education radically fair — giving every student, regardless of family income or geography, a real chance to feel the same wonder at the natural world that set me on this path.
UTRIP 2025, University of Tokyo
GAT-based particle position reconstruction for dark matter direct detection (undergraduate thesis). Developing an ML-driven detector health monitoring system for 24/7 anomaly diagnosis.
Developed and characterized novel water-based liquid scintillator for next-generation neutrino detectors; evaluated comparative PMT charge response against pure water. Lab is a core group of the Super-Kamiokande experiment (Nobel Prize in Physics 2015, neutrino oscillation).
ML-based particle energy and position reconstruction for the Water Cherenkov Test Experiment; explored model inversion and interpretability techniques.
ML-based particle position reconstruction and detector signal classification for neutrino detection experiments.
Developing benchmark datasets in particle physics and astrophysics.
Built end-to-end training pipelines for on-device LLMs: dataset construction, SFT with prompt engineering, LoRA and full-parameter fine-tuning across dense and MoE architectures, and GRPO post-training alignment. Participated in an internal hackathon to incubate an agent product based on the OpenClaw architecture.