Henry Day-Hall

I build generative models that stand in for slow physics simulation, and I get them into production. Postdoctoral researcher at DESY in Hamburg.

Download CV Email me

About

I design and train generative models that replace slow Monte Carlo simulation. At DESY I lead work on surrogate models for the calorimeters of future Higgs factories: CaloClouds 3 generates photon showers around a hundred times faster than standard simulation, and AllShowers covers twelve particle species in a single model — the first time that has been done. Most of my interest is in what happens after the paper: distilling models down for inference, compiling them with TorchScript, and getting them into the C++ toolchains where other people can actually use them.

Before Hamburg I validated fast simulation for ATLAS at CERN, working inside a seven-million-line codebase maintained by developers in forty countries, and worked on jet and dijet cross-sections. My PhD was on jet clustering built from eigenvalue spectra. On the side I am tracking particle histories through extensive air showers with CORSIKA 8.

Henry Day-Hall

Work

  • 2024—now

    Postdoctoral researcher · DESY, Hamburg

    Generative surrogate simulation for future collider experiments. Led CaloClouds 3 and contributed to AllShowers; distilled models for roughly hundred-fold faster inference; used transfer learning to carry models between detector geometries and materials. Supervise summer and masters students.

  • 2021—2023

    Postdoctoral researcher · Czech Technical University, Prague

    Validation and deployment of fast simulation for ATLAS (AtlFast3 and FastCaloSim v3). Automated the validation pipeline while keeping it flexible enough for a stream of new models, and refactored the C++ production environment to support several inference runtimes. Member of the collaboration during the period recognised by the 2025 Breakthrough Prize in Fundamental Physics.

  • 2017—2021

    PhD, machine learning for high energy physics · University of Southampton

    Developed spectral clustering for jet formation, supervised by an interdisciplinary team spanning physics and computer science. Trained in the NGCM and NExT doctoral centres. Seconded to Cadi Ayyad University in Morocco, where I profiled and refactored Higgs-physics software for a 30% speed-up and containerised its stack.

  • 2012—2017

    MSc theoretical physics · University of Edinburgh

    First class integrated masters, thesis on QCD on the light cone. Research projects with the Hyper-Kamiokande collaboration and the theory group at TU Munich.

Projects

Publications

Talks and outreach

  • 2025

    8th Round Table on Deep Learning, DESY

    Invited keynote on large language models for users.

  • 2024

    City of Science Day, DESY

    Outreach stall on the fundamentals of neural networks and where particle physics uses them.

  • 2022

    Efficient simulations on GPU hardware, ETH Zürich

    Talk on architecture and library choices for getting the most out of available GPU resources in generative detector simulation.

  • 2022

    ATLAS Week, University of Lisbon

    Poster on the validation toolkit for FastCaloSim 3.

  • 2022

    Workshop JČF, Bílý Potok

    Seminar on design patterns and building physics code that survives contact with other people.

  • 2021

    Seminar, KEK, Tokyo

    Invited talk on spectral clustering for jet formation, to Prof. Mihoko Nojiri's working group.

  • 2019

    1st Mediterranean Conference on Higgs Physics, Tangier

    Talk on determining parameter spaces using container stacks without administrative privileges.

  • 2019

    Science and Engineering Festival, University of Southampton

    Bench-top experiments and interactive performances explaining how a particle collider works.

  • 2014

    TEDx, University of Edinburgh

    On the surprising accuracy of Wikipedia, and how to use it in unfamiliar territory. Watch the talk.

Service and supervision

  • Joint ML Journal Club — founder and coordinator of a club where DESY and Universität Hamburg read the latest machine learning work together, including plenty from outside physics.
  • Early Career Scientist Boards — representative at DESY FH since 2024, and at ATLAS from 2022 to 2023. Surveys, peer mentoring, welcome days and carrying early-career concerns to management.
  • Supervision — DESY summer students Jemma Bagg (2026, continuous normalising flows for inference speed), Aysu Kocaalp (2025, inverting fast simulation models to build reconstruction tools) and Simon Edgar Pijahn (2024, evaluating fast simulation with discriminator networks). Also a contributor to the supervison of PhD and Master students in FTX-SFT.
  • Organisation — FTX seminar series. FTX summer student program. Conferences and workshops at DESY and at CVUT Prague.