Dr Benjamin Hodkinson

Abstract:

Monte Carlo simulations are a crucial component when analysing the Standard Model and New physics processes at the Large Hadron Collider (LHC). This paper aims to explore the use of generative models for increasing the statistics of Monte Carlo simulations in the final stage of data analysis by generating synthetic data that follows the same kinematic distributions for a limited set of analysis-specific observables to a high precision. Several state-of-the-art generative machine learning algorithms are adapted to this task, best performance is demonstrated by the normalizing flow architectures, which are capable of fast generation of an arbitrary number of new events. As an example of analysis-specific Monte Carlo simulated data, a well-known benchmark sample containing the Higgs boson production beyond the Standard Model and the corresponding irreducible background is used. The applicability of normalizing flows with different model parameters and numbers of initial events used in training is investigated. The resulting event distributions are compared with the original Monte Carlo distributions using statistical tests and a simplified statistical analysis to evaluate their similarity and quality of reproduction required in a physics analysis environment in a systematic way.Comment: 28 pages, 22 figure

Abstract:

We consider a model where the electroweak symmetry breaking is driven by strong dynamics, resulting in an electroweak doublet scalar condensate, and transmitted to the standard model matter fields via another electroweak doublet scalar. At low energies the effective theory therefore shares features with a type-I two Higgs doublet model. However, important differences arise due to the rich composite spectrum expected to contain new vector resonances accessible at the LHC. We carry out a systematic analysis of the vector resonance signals at LHC and find that the model remains viable, but will be tightly constrained by direct searches as the projected integrated luminosity, around 200 fb(-1), of the current run becomes available.Peer reviewe

Abstract:

This paper presents a measurement of the electroweak production of two jets in association with a Zγ pair, with the Z boson decaying into two neutrinos. It also presents a search for invisible or partially invisible decays of a Higgs boson with a mass of 125 GeV produced through vectorboson fusion with a photon in the final state. These results use data from LHC proton–proton collisions at sqrt(s) = 13 TeV collected with the ATLAS detector and corresponding to an integrated luminosity of 139 fb−1. The event signature, shared by all benchmark processes considered for the measurements and searches, is characterized by a significant amount of unbalanced transverse momentum and a photon in the final state, in addition to a pair of forward jets. Electroweak Zγ production in association with two jets is observed in this final state with a significance of 5.2 (5.1 expected) standard deviations. The measured fiducial cross-section for this process is 1.31 ± 0.29 fb. An observed (expected) upper limit of 0.37 (0.34+0.15 −0.10) at 95% confidence level is set on the branching ratio of a 125 GeV Higgs boson to invisible particles, assuming the Standard Model production cross-section. The signature is also interpreted in the context of decays of a Higgs boson into a photon and a dark photon. An observed (expected) 95% CL upper limit on the branching ratio for this decay is set at 0.018 (0.017+0.007−0.005), assuming the Standard Model production cross-section for a 125 GeV Higgs boson

Abstract:

Deep learning is a standard tool in the field of high-energy physics, facilitating considerable sensitivity enhancements for numerous analysis strategies. In particular, in identification of physics objects, such as jet flavor tagging, complex neural network architectures play a major role. However, these methods are reliant on accurate simulations. Mismodeling can lead to non-negligible differences in performance in data that need to be measured and calibrated against. We investigate the classifier response to input data with injected mismodelings and probe the vulnerability of flavor tagging algorithms via application of adversarial attacks. Subsequently, we present an adversarial training strategy that mitigates the impact of such simulated attacks and improves the classifier robustness. We examine the relationship between performance and vulnerability and show that this method constitutes a promising approach to reduce the vulnerability to poor modeling.Comment: 17 pages, 16 figures, 2 tables. Replaced with the published version. Added the journal reference and the DOI. Code accessible under https://github.com/AnnikaStein/Adversarial-Training-for-Jet-Taggin