Searcharxiv⌕ Search

arXiv subjects

Stefano Carrazza

Publications and source records attributed to Stefano Carrazza.

At least 73 records · Page 4Linked to original sources
arXiv2019

Towards hardware acceleration for parton densities estimation

In this proceedings we describe the computational challenges associated to the determination of parton distribution functions (PDFs). We compare the performance of the convolution of the parton distributions with matrix elements using different hardware instructions. We quantify and identify the most promising data-model configurations to increase PDF fitting performance in adapting the current code frameworks to hardware accelerators such as graphics processing units.

hep-ph↗
arXiv2019

libGroomRL: Reinforcement Learning for Jets

In these proceedings, we present a library allowing for straightforward calls in C++ to jet grooming algorithms trained with deep reinforcement learning. The RL agent is trained with a reward function constructed to optimize the groomed jet properties, using both signal and background samples in a simultaneous multi-level training. We show that the grooming algorithm derived from the deep RL agent can match state-of-the-art techniques used at the Large Hadron Collider, resulting in improved mass resolution for boosted objects. Given a suitable reward function, the agent learns how to train a policy which optimally removes soft wide-angle radiation, allowing for a modular grooming technique that can be applied in a wide range of contexts. The neural network trained with GroomRL can be used in a FastJet analysis through the libGroomRL C++ library.

physics.data-an↗
arXiv2019

Jet grooming through reinforcement learning

We introduce a novel implementation of a reinforcement learning (RL) algorithm which is designed to find an optimal jet grooming strategy, a critical tool for collider experiments. The RL agent is trained with a reward function constructed to optimize the resulting jet properties, using both signal and background samples in a simultaneous multi-level training. We show that the grooming algorithm derived from the deep RL agent can match state-of-the-art techniques used at the Large Hadron Collider, resulting in improved mass resolution for boosted objects. Given a suitable reward function, the agent learns how to train a policy which optimally removes soft wide-angle radiation, allowing for a modular grooming technique that can be applied in a wide range of contexts. These results are accessible through the corresponding GroomRL framework.

hep-ph↗
arXiv2019

Towards a new generation of parton densities with deep learning models

We present a new regression model for the determination of parton distribution functions (PDF) using techniques inspired from deep learning projects. In the context of the NNPDF methodology, we implement a new efficient computing framework based on graph generated models for PDF parametrization and gradient descent optimization. The best model configuration is derived from a robust cross-validation mechanism through a hyperparametrization tune procedure. We show that results provided by this new framework outperforms the current state-of-the-art PDF fitting methodology in terms of best model selection and computational resources usage.

hep-ph↗
arXiv2019

Modelling conditional probabilities with Riemann-Theta Boltzmann Machines

The probability density function for the visible sector of a Riemann-Theta Boltzmann machine can be taken conditional on a subset of the visible units. We derive that the corresponding conditional density function is given by a reparameterization of the Riemann-Theta Boltzmann machine modelling the original probability density function. Therefore the conditional densities can be directly inferred from the Riemann-Theta Boltzmann machine.

stat.ML↗
arXiv2019

Machine Learning in High Energy Physics Community White Paper

Machine learning has been applied to several problems in particle physics research, beginning with applications to high-level physics analysis in the 1990s and 2000s, followed by an explosion of applications in particle and event identification and reconstruction in the 2010s. In this document we discuss promising future research and development areas for machine learning in particle physics. We detail a roadmap for their implementation, software and hardware resource requirements, collaborative initiatives with the data science community, academia and industry, and training the particle physics community in data science. The main objective of the document is to connect and motivate these areas of research and development with the physics drivers of the High-Luminosity Large Hadron Collider and future neutrino experiments and identify the resource needs for their implementation. Additionally we identify areas where collaboration with external communities will be of great benefit.

physics.comp-ph↗
arXiv2019

Can New Physics hide inside the proton?

Modern global analyses of the structure of the proton include collider measurements which probe energies well above the electroweak scale. While these provide powerful constraints on the parton distribution functions (PDFs), they are also sensitive to beyond the Standard Model (BSM) dynamics if these affect the fitted distributions. Here we present a first simultaneous determination of the PDFs and BSM effects from deep-inelastic structure function data by means of the NNPDF framework. We consider representative four-fermion operators from the SM Effective Field Theory (SMEFT), quantify to which extent their effects modify the fitted PDFs, and assess how the resulting bounds on the SMEFT degrees of freedom are modified. Our results demonstrate how BSM effects that might otherwise be reabsorbed into the PDFs can be systematically disentangled.

hep-ph↗
arXiv2018

MINLO t-channel single-top plus jet

We present a next-to-leading order accurate simulation of t-channel single-top plus jet production matched to parton showers via the POWHEG method. The calculation underlying the simulation is enhanced with a process-specific implementation of the multi-scale improved NLO (MINLO) method, such that it gives physical predictions all through phase space, including regions where the jet additional to the t-channel single-top process is unresolved. We further describe a tuning procedure for the MINLO Sudakov form factor, fitting the coefficient of the first subleading term in its exponent using an artificial neural-network. The latter tuning, implemented as a straightforward event-by-event reweighting, renders the MINLO simulation NLO accurate for t-channel single-top observables, in addition to those of the analogous single-top plus jet process.

hep-ph↗
arXiv2018

Precision determination of the strong coupling constant within a global PDF analysis

We present a determination of the strong coupling constant $α_s(m_Z)$ based on the NNPDF3.1 determination of parton distributions, which for the first time includes constraints from jet production, top-quark pair differential distributions, and the $Z$ $p_T$ distributions using exact NNLO theory. Our result is based on a novel extension of the NNPDF methodology - the correlated replica method - which allows for a simultaneous determination of $α_s$ and the PDFs with all correlations between them fully taken into account. We study in detail all relevant sources of experimental, methodological and theoretical uncertainty. At NNLO we find $α_s(m_Z) = 0.1185 \pm 0.0005^\text{(exp)}\pm 0.0001^\text{(meth)}$, showing that methodological uncertainties are negligible. We conservatively estimate the theoretical uncertainty due to missing higher order QCD corrections (N$^3$LO and beyond) from half the shift between the NLO and NNLO $α_s$ values, finding $Δα^{\rm th}_s =0.0011$.

hep-ph↗
arXiv2018

Illuminating the photon content of the proton within a global PDF analysis

Precision phenomenology at the LHC requires accounting for both higher-order QCD and electroweak corrections as well as for photon-initiated subprocesses. Building upon the recent NNPDF3.1 fit, in this work the photon content of the proton is determined within a global analysis supplemented by the LUXqed constraint relating the photon PDF to lepton-proton scattering structure functions: NNPDF3.1luxQED. The uncertainties on the resulting photon PDF are at the level of a few percent, with photons carrying up to 0.5% of the proton's momentum. We study the phenomenological implications of NNPDF3.1luxQED at the LHC for Drell-Yan, vector boson pair, top quark pair, and Higgs plus vector boson production. We find that photon-initiated contributions can be significant for many processes, leading to corrections of up to 20%. Our results represent a state-of-the-art determination of the partonic structure of the proton including its photon component.

hep-ph↗
arXiv2018

Machine learning challenges in theoretical HEP

In these proceedings we perform a brief review of machine learning (ML) applications in theoretical High Energy Physics (HEP-TH). We start the discussion by defining and then classifying machine learning tasks in theoretical HEP. We then discuss some of the most popular and recent published approaches with focus on a relevant case study topic: the determination of parton distribution functions (PDFs) and related tools. Finally, we provide an outlook about future applications and developments due to the synergy between ML and HEP-TH.

hep-ph↗
arXiv2017

Minimisation strategies for the determination of parton density functions

We discuss the current minimisation strategies adopted by research projects involving the determination of parton distribution functions (PDFs) and fragmentation functions (FFs) through the training of neural networks. We present a short overview of a proton PDF determination obtained using the covariance matrix adaptation evolution strategy (CMA-ES) optimisation algorithm. We perform comparisons between the CMA-ES and the standard nodal genetic algorithm (NGA) adopted by the NNPDF collaboration.

hep-ph↗
arXiv2017

Parton distributions from high-precision collider data

We present a new set of parton distributions, NNPDF3.1, which updates NNPDF3.0, the first global set of PDFs determined using a methodology validated by a closure test. The update is motivated by recent progress in methodology and available data, and involves both. On the methodological side, we now parametrize and determine the charm PDF alongside the light quarks and gluon ones, thereby increasing from seven to eight the number of independent PDFs. On the data side, we now include the D0 electron and muon W asymmetries from the final Tevatron dataset, the complete LHCb measurements of W and Z production in the forward region at 7 and 8 TeV, and new ATLAS and CMS measurements of inclusive jet and electroweak boson production. We also include for the first time top-quark pair differential distributions and the transverse momentum of the Z bosons from ATLAS and CMS. We investigate the impact of parametrizing charm and provide evidence that the accuracy and stability of the PDFs are thereby improved. We study the impact of the new data by producing a variety of determinations based on reduced datasets. We find that both improvements have a significant impact on the PDFs, with some substantial reductions in uncertainties, but with the new PDFs generally in agreement with the previous set at the one sigma level. The most significant changes are seen in the light-quark flavor separation, and in increased precision in the determination of the gluon. We explore the implications of NNPDF3.1 for LHC phenomenology at Run II, compare with recent LHC measurements at 13 TeV, provide updated predictions for Higgs production cross-sections and discuss the strangeness and charm content of the proton in light of our improved dataset and methodology. The NNPDF3.1 PDFs are delivered for the first time both as Hessian sets, and as optimized Monte Carlo sets with a compressed number of replicas.

hep-ph↗
arXiv2017

A determination of the fragmentation functions of pions, kaons, and protons with faithful uncertainties

We present NNFF1.0, a new determination of the fragmentation functions (FFs) of charged pions, charged kaons, and protons/antiprotons from an analysis of single-inclusive hadron production data in electron-positron annihilation. This determination, performed at leading, next-to-leading, and next-to-next-to-leading order in perturbative QCD, is based on the NNPDF methodology, a fitting framework designed to provide a statistically sound representation of FF uncertainties and to minimise any procedural bias. We discuss novel aspects of the methodology used in this analysis, namely an optimised parametrisation of FFs and a more efficient $χ^2$ minimisation strategy, and validate the FF fitting procedure by means of closure tests. We then present the NNFF1.0 sets, and discuss their fit quality, their perturbative convergence, and their stability upon variations of the kinematic cuts and the fitted dataset. We find that the systematic inclusion of higher-order QCD corrections significantly improves the description of the data, especially in the small-$z$ region. We compare the NNFF1.0 sets to other recent sets of FFs, finding in general a reasonable agreement, but also important differences. Together with existing sets of unpolarised and polarised parton distribution functions (PDFs), FFs and PDFs are now available from a common fitting framework for the first time.

hep-ph↗
arXiv2017

Modeling NNLO jet corrections with neural networks

We present a preliminary strategy for modeling multidimensional distributions through neural networks. We study the efficiency of the proposed strategy by considering as input data the two-dimensional next-to-next leading order (NNLO) jet k-factors distribution for the ATLAS 7 TeV 2011 data. We then validate the neural network model in terms of interpolation and prediction quality by comparing its results to alternative models.

hep-ph↗
arXiv2016

A Determination of the Charm Content of the Proton

We present an unbiased determination of the charm content of the proton, in which the charm parton distribution function (PDF) is parametrized on the same footing as the light quarks and the gluon in a global PDF analysis. This determination relies on the calculation of deep-inelastic structure functions in the FONLL scheme, generalized to account for massive charm-initiated contributions. In contrast to the usual situation in which the charm PDF is assumed to be generated perturbatively by pair radiation off gluons and light quarks, vanishing at a scale set by the value of the charm mass m_c, we find that the fitted charm PDF vanishes within uncertainties at a scale Q~1.5 GeV for all x<~0.1, independent of the value of m_c used in the coefficient functions. We also find some evidence that the charm PDF at large x>~0.1 and low scales does not vanish, but rather has an "intrinsic" component, very weakly scale dependent and almost independent of the value of m_c, carrying about 1% of the total momentum of the proton. The uncertainties in all other PDFs are only slightly increased by the inclusion of fitted charm, while the dependence of these PDFs on m_c is significantly reduced. When the EMC charm structure function dataset is included, it is well described by the fit, and PDF uncertainties in the fitted charm PDF are significantly reduced, though we verify that excluding the EMC data does not qualitatively modify any of our findings. The increased stability with respect to m_c persists at high scales and is the main implication of our results for LHC phenomenology. Fitting the charm PDF modifies the predictions for processes such as high p_T and large rapidity charm pair production and Z+c production, and thus we expect that future LHC data will further constrain the charm content of the proton.

hep-ph↗
arXiv2016

QCDLoop: a comprehensive framework for one-loop scalar integrals

We present a new release of the QCDLoop library based on a modern object-oriented framework. We discuss the available new features such as the extension to the complex masses, the possibility to perform computations in double and quadruple precision simultaneously, and useful caching mechanisms to improve the computational speed. We benchmark the performance of the new library, and provide practical examples of phenomenological implementations by interfacing this new library to Monte Carlo programs.

hep-ph↗