I am Eugene (Yoogeun) Song, a PhD researcher in high-energy physics at Imperial College London, working on neutrinos with DUNE and NOvA under Dr Linda Cremonesi. Physical modelling, machine learning and statistical inference are not three disciplines to me. They are one problem with three names: inference under uncertainty.

Currently working on, from 2026 onwards:

• 🌌 Neutrino oscillations — systematics-aware ML reconstruction, Bayesian and MCMC inference, and Near-to-Far Detector constraint propagation for the DUNE Near Detector, alongside a variety of physics analyses on NOvA, which has data today
• 🫀 Physics-informed neural networks — real-time atrial fibrillation mapping from grid electrograms, at ~78 ms latency and ~1.6 mm RMS localisation error
• ⚛️ Quantum-accelerated CFD — the classical solver and hybrid layer behind non-invasive cardiovascular diagnostics
• 📈 Quantitative research — alpha generation under non-stationary market dynamics

Previously: general-relativistic magnetohydrodynamics of Sagittarius A*, black hole magnetospheres and Blandford–Znajek energy extraction, and early-universe cosmology.

I enrolled in university physics at age 7 and finished a BSc at age 11, a record that remains unmatched in Korea. My first first-author paper followed at age 19, in MNRAS Letters, on gamma radiation from extremely rotating black holes. Twenty years on I am still chasing hard physics problems, neutrinos now. The early start is the part people remember; the part I care about is that the questions I need to ask, and to answer, have only got harder over time. Finding the right question to ask about the universe is most of the work, and answering it properly is the rest.

Python · PyTorch · NumPy / SciPy · C++ · Fortran · Julia · Mathematica · CUDA · MPI · LaTeX · Git

Machine learning should augment physical interpretation, never replace it. A network that improves resolution while hiding its own failure modes is a worse instrument than a slower method whose biases you can enumerate. Human intuition and qualitative judgement are not the soft part of the process; they are what decides whether a number means anything. Used with that much care, and only then, machine learning is among the most powerful tools humans have ever built.