A5076449537- Particle physics theoretical and experimental studies
- Computational Physics and Python Applications
- High-Energy Particle Collisions Research
- Distributed and Parallel Computing Systems
- Particle Detector Development and Performance
- Quantum Chromodynamics and Particle Interactions
- Cosmology and Gravitation Theories
- Dark Matter and Cosmic Phenomena
- Scientific Computing and Data Management
- Black Holes and Theoretical Physics
- Astrophysics and Cosmic Phenomena
- Gaussian Processes and Bayesian Inference
- Model Reduction and Neural Networks
- Reservoir Engineering and Simulation Methods
- Neural Networks and Applications
- Algorithms and Data Compression
- Soil and Unsaturated Flow
- Medical Imaging Techniques and Applications
- Radiation Detection and Scintillator Technologies
- Numerical Methods and Algorithms
- Electromagnetic Simulation and Numerical Methods
- Gamma-ray bursts and supernovae
- Distributed Sensor Networks and Detection Algorithms
- Radiation Effects in Electronics
- Pulsars and Gravitational Waves Research
Université Paris Cité
2022-2025
Sorbonne University Abu Dhabi
2022-2025
Sorbonne Université
2022-2025
Heidelberg University
2016-2025
Institut National de Physique Nucléaire et de Physique des Particules
2016-2025
Laboratoire de Physique Nucléaire et de Hautes Énergies
2022-2025
Centre National de la Recherche Scientifique
2016-2024
Sorbonne Paris Cité
2024
Laboratoire de Physique Théorique et Hautes Energies
2024
Université Paris 8
2023
A critical question concerning generative networks applied to event generation in particle physics is if the generated events add statistical precision beyond training sample. We show for a simple example with increasing dimensionality how indeed amplify statistics. quantify their impact through an amplification factor or equivalent numbers of sampled events.
First-principle simulations are at the heart of high-energy physics research program. They link vast data output multi-purpose detectors with fundamental theory predictions and interpretation. This review illustrates a wide range applications modern machine learning to event generation simulation-based inference, including conceptional developments driven by specific requirements particle physics. New ideas tools developed interface will improve speed precision forward simulations, handle...
Generative networks are opening new avenues in fast event generation for the LHC. We show how generative flow can reach percent-level precision kinematic distributions, they be trained jointly with a discriminator, and this discriminator improves generation. Our joint training relies on novel coupling of two which does not require Nash equilibrium. then estimate uncertainties through Bayesian network setup conditional data augmentation, while ensures that there no systematic inconsistencies...
We provide an overview of the status Monte-Carlo event generators for high-energy particle physics. Guided by experimental needs and requirements, we highlight areas active development, opportunities future improvements. Particular emphasis is given to physics models algorithms that are employed across a variety experiments. These common themes in generator development lead more comprehensive understanding at highest energies intensities, allow be tested against wealth data have been...
We introduce two diffusion models and an autoregressive transformer for LHC physics simulations. Bayesian versions allow us to control the networks capture training uncertainties. After illustrating their different density estimation methods simple toy models, we discuss advantages Z plus jets event generation. While excel through precision, scales best with phase space dimensionality. Given evaluation speed, expect benefit from dedicated use cases normalizing flows, transformers.
For simulations where the forward and inverse directions have a physics meaning, invertible neural networks are especially useful. A conditional INN can invert detector simulation in terms of high-level observables, specifically for ZW production at LHC. It allows per-event statistical interpretation. Next, we allow variable number QCD jets. We unfold effects radiation to pre-defined hard process, again with probabilistic interpretation over parton-level phase space.
LHC analyses directly comparing data and simulated events bear the danger of using first-principle predictions only as a black-box part event simulation. We show how simulations, for instance, detector effects can instead be inverted generative networks. This allows us to reconstruct parton level information from measured events. Our results illustrate how, in general, fully conditional networks statistically invert Monte Carlo simulations. As technical by-product we maximum mean discrepancy...
The effective Lagrangian expansion provides a framework to study effects of new physics at the electroweak scale. To make full use LHC data in constraining higher-dimensional operators we need include both Higgs and gauge sector our study. We first present an analysis relevant di-boson production results update constraints on triple boson couplings. Our bounds are several times stronger than those obtained from LEP data. Next, show how combination with measurements vertices lead significant...
Based on the established task of identifying boosted, hadronically decaying top quarks, we compare a wide range modern machine learning approaches. Unlike most methods they rely low-level input, for instance calorimeter output. While their network architectures are vastly different, performance is comparatively similar. In general, find that these new approaches extremely powerful and great fun.
We introduce a new and highly efficient tagger for hadronically decaying top quarks, based on deep neural network working with Lorentz vectors the Minkowski metric. With its novel machine learning setup architecture it allows us to identify boosted quarks not only from calorimeter towers, but also including tracking information. show how performance of our compares QCD-inspired image-recognition approaches find that significantly increases strongly quarks.
QCD splittings are among the most fundamental theory concepts at LHC. We show how they can be studied systematically with help of invertible neural networks. These networks work sub-jet information to extract parameters from jet samples. Our approach expands LEP measurements Casimirs a systematic test properties based on low-level observables. Starting an toy example we study effect full shower, hadronization, and detector effects in detail.
Event generation with neural networks has seen significant progress recently. The big open question is still how such new methods will accelerate LHC simulations to the level required by upcoming runs. We target a known bottleneck of standard and show their unweighting procedure can be improved generative networks. This can, potentially, lead very gain in simulation speed.
Numerical evaluations of Feynman integrals often proceed via a deformation the integration contour into complex plane. While valid contours are easy to construct, numerical precision for multi-loop integral can depend critically on chosen contour. We present methods optimize this using combination optimized, global shifts and normalizing flow. They lead significant gain in precision.
Abstract Motivated by the high computational costs of classical simulations, machine-learned generative models can be extremely useful in particle physics and elsewhere. They become especially attractive when surrogate efficiently learn underlying distribution, such that a generated sample outperforms training limited size. This kind GANplification has been observed for simple Gaussian models. We show same effect simulation, specifically photon showers an electromagnetic calorimeter.
The matrix element method is widely considered the ultimate LHC inference tool for small event numbers. We show how a combination of two conditional generative neural networks encodes QCD radiation and detector effects without any simplifying assumptions, while keeping computation likelihoods individual events numerically efficient. illustrate our approach CP-violating phase top Yukawa coupling in associated Higgs single-top production. Currently, limiting factor precision jet combinatorics.
We introduce two diffusion models and an autoregressive transformer for LHC physics simulations. Bayesian versions allow us to control the networks capture training uncertainties. After illustrating their different density estimation methods simple toy models, we discuss advantages Z plus jets event generation. While excel through precision, scales best with phase space dimensionality. Given evaluation speed, expect benefit from dedicated use cases normalizing flows, transformers.
Evaluating loop amplitudes is a time-consuming part of LHC event generation. For di-photon production with jets we show that simple, Bayesian networks can learn such and model their uncertainties reliably. A boosted training the network further improves uncertainty estimate precision in critical phase space regions. In general, allows us to move between fit-like interpolation-like regimes training.
Event generation for the LHC can be supplemented by generative adversarial networks, which generate physical events and avoid highly inefficient event unweighting. For top pair production we show how such a network describes intermediate on-shell particles, phase space boundaries, tails of distributions. In particular, introduce maximum mean discrepancy to resolve sharp local features. It extended in straightforward manner include instance off-shell contributions, higher orders, or...
Subtracting event samples is a common task in LHC simulation and analysis, standard solutions tend to be inefficient. We employ generative adversarial networks produce new with phase space distribution corresponding added or subtracted input samples. first illustrate for toy example how such network beats the statistical limitations of training data. then show can used subtract background events include non-local collinear subtraction at level unweighted 4-vector events.
Abstract Machine learning tools have empowered a qualitatively new way to perform differential cross section measurements whereby the data are unbinned, possibly in many dimensions. Unbinned can enable, improve, or at least simplify comparisons between experiments and with theoretical predictions. Furthermore, many-dimensional be used define observables after measurement instead of before. There is currently no community standard for publishing unbinned data. While there also essentially...
QCD-jets at the LHC are described by simple physics principles. We show how super-resolution generative networks can learn underlying structures and use them to improve resolution of jet images. test this approach on massless fat top-jets find that network reproduces their main features even without training pure samples. In addition, we a slim architecture be constructed once have control full performance.
Theory predictions for the LHC require precise numerical phase-space integration and generation of unweighted events. We combine machine-learned multi-channel weights with a normalizing flow importance sampling, to improve classical methods integration. develop an efficient bi-directional setup based on invertible network, combining online buffered training potentially expensive integrands. illustrate our method Drell-Yan process additional narrow resonance.
While neural networks offer an attractive way to numerically encode functions, actual formulas remain the language of theoretical particle physics. We use symbolic regression trained on matrix-element information extract, for instance, optimal LHC observables. This we invert usual simulation paradigm and extract easily interpretable from complex simulated data. introduce method using effect a dimension-6 coefficient associated ZH production. then validate it known case CP-violation in...
Off-shell effects in large LHC backgrounds are crucial for precision predictions and, at the same time, challenging to simulate. We present a novel method transform high-dimensional distributions based on diffusion neural network and use it generate process with off-shell kinematics from much simpler on-shell one. Applied toy example of top pair production LO we show how our generates configurations fast precisely, while reproducing even features.