Leverhulme lecture - Testing the MiniBooNE Anomaly with MicroBooNE, and Building the ML Toolkit for What Comes Next

by Taritree Wongjirad

Friday 8 May 2026, 13:00 → 14:00 Europe/London

Bohr Conference Room

The MicroBooNE experiment, an 85-tonne liquid argon time projection chamber on the Booster Neutrino Beam at Fermilab, was built to test the long-standing MiniBooNE low-energy excess. I will present two recent MicroBooNE results that together address both leading interpretations of that anomaly. Using a novel two-beam analysis combining BNB and NuMI data, MicroBooNE excludes the single light sterile neutrino interpretation of the LSND and MiniBooNE anomalies at 95% CL, with strong consistency with the three-neutrino hypothesis. A complementary inclusive single-photon search probes the photon-induced interpretation, and finds a ~2σ excess of low-energy events in the zero-proton sub-sample. This is an intriguing hint, but one obtained from a selection with only ~7% efficiency and ~40% purity, illustrating how challenging photon-channel searches are with current reconstruction tools.

This challenge motivates the second half of the talk: a survey of recent machine-learning developments — including deep-learning reconstruction now demonstrated on real MicroBooNE data, self-supervised foundation models for LArTPC images, generative and differentiable detector simulation, and ML-based unfolding — that promise substantial gains in efficiency, purity, and analysis flexibility. Several of these tools are being developed in part during my ongoing sabbatical here at the University of Manchester. They will be applied to the remaining MicroBooNE data and form a real-data test bed for the next generation of LArTPC experiments, including SBN and DUNE.