Nick Haubrich

Higgs 2022 in Pisa

I attended the Higgs 2022 conference in Pisa, Italy, and gave a talk on my analysis (VHbb) and related measurements from CMS. My slides are publicly available here. The conference chatter generally revolved around the success of graph neural networks for jet tagging and regression (like in the HH4b analysis I presented) and the constant headache of theory uncertainties.

Matthias Kerner’s talk showing cross-section differences of 2x-20x in ggZH at NLO vs LO (see this slide) was particularly troubling. In VHbb, our leading-order understanding has been qqZH dominates the ggZH contribution, but that may not be the case in the boosted regime.

Disagreement between parton shower algorithms is becoming relevant for more analyses as statistical errors shrink. ATLAS’s two-point comparisons leave a lot to be desired but there’s value in having a consistent approach. In CMS the strategy is to avoid depence on such differences then trust one algorithm, and that seems to be born from a lack of simulated samples compared to ATLAS. If only we could trade ParticleNet for a few billion SHERPA events from ATLAS. Personally I find the parton shower differences concerning for these graph neural networks that are exposed to such low-level observables. I’m skeptical that we’d be able to correct biases arising from the parton shower with the standard jet corrections or even some adversarial training approach.

There was a bit of time for sightseeing too.