Speaker
Description
"The Standard Model (SM) describes the known elementary particles with high precision, yet it does not explain the origin of the electroweak scale, incorporate gravity, or provide a dark-matter candidate — shortcomings motivating the search for physics beyond the Standard Model (BSM). Supersymmetry (SUSY) is a leading BSM candidate; in its gauge-mediated supersymmetry breaking (GMSB) branch, SUSY breaking is transmitted by gauge interactions and the lightest supersymmetric particle is a very light gravitino. The resulting compressed spectra yield soft final states and modest missing transverse energy, a regime where prompt searches lose sensitivity.
Exploring this space, however, relies mostly on low-dimensional simplified models, which give exclusion contours of low-dimensional projection but cannot represent a realistic high-dimensional space, where a point is rarely described by one topology and its sensitivity spreads across analyses. Under flat-prior (uniform random) sampling, such a space suffers a severe curse of dimensionality: viable and informative regions occupy a tiny fraction of it, so most of the budget goes to points that are deeply excluded or allowed, while the exclusion boundary remains sparsely sampled.
To close this gap, we have applied active learning (AL) to the nine-dimensional GMSB parameter space, where at equal budget it markedly improves the sampling efficiency near the exclusion contour relative to flat-prior sampling and standard MCMC. Building on this, we are extending the framework towards global coverage in three directions: a calibrated multi-level reinterpretation chain with explicit escalation criteria; a cost-aware acquisition strategy that jointly selects model points and evaluation levels; and a robustness assessment that retains only reproducibly recovered weak regions as candidate blind spots, turning them into benchmarks for future LHC searches."
