Zichun Hao
Ph.D. Candidate in Physics, California Institute of Technology
I am a physics Ph.D. candidate at Caltech advised by Prof. Maria Spiropulu, working at the intersection of machine learning and particle physics.
My research develops self-supervised foundation models for collider data. I lead RINO, a spotlight paper at the ML4PS workshop at NeurIPS 2025 that pretrains transformers on unlabeled jets with DINO-style self-distillation and improves out-of-distribution generalization over fully-supervised training. During my undergraduate years I did research in the Duarte Lab, mentored by Prof. Javier Duarte and Dr. Raghav Kansal. There I built Lorentz group equivariant autoencoders, published in The European Physical Journal C, developed a tagger for boosted Higgs boson decays to W boson pairs, and contributed to JetNet, an open-source library for machine learning in high energy physics.
At the CMS experiment at CERN, I am one of the main contributors to the boosted Higgs boson pair search in the four bottom quark final state, working across the analysis chain from tagger calibration to the final fits. I also enjoy teaching and co-designed the reinforcement learning final project on quantum state preparation for Caltech’s Ph 22, Computational Physics Laboratory III.
News
| Jun 25, 2026 | Our paper High Multiplicity Trigger for Long-Lived Particles in CMS Detector is published in JINST. The paper presents a dedicated trigger that targets unusually large hit multiplicities in the CMS muon detectors, a characteristic signature of long-lived particles decaying in the muon system, substantially extending the CMS sensitivity to such signatures during LHC Run 3. I contributed the electron+MDS and muon+MDS cross triggers, which were accepted into the CMS trigger menu in 2023. |
|---|---|
| Apr 30, 2026 | The CMS analysis I work on, Improved results on Higgs boson pair production in the 4b final state, is now public and submitted to Physical Review D. It sets the most stringent limit on non-resonant $\mathrm{HH}$ production to date. |
| Dec 06, 2025 | RINO was selected as a ML4PS workshop spotlight at NeurIPS 2025, an honor offered to about 1% of the workshop’s accepted papers. I presented the spotlight talk and poster in San Diego. |