GANplifying Event Samples

Butter, Anja; Diefenbacher, Sascha; Kasieczka, Gregor; Nachman, Benjamin; Plehn, Tilman

doi:10.48550/arxiv.2008.06545

Cited by 8 publications

(13 citation statements)

References 47 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For the LHC, this can be seen for instance in the non-trivial uncertainty for an intermediate Breit-Wigner resonance. These results are another step in understanding GANplification patterns [4] and might even allow us to use INNs to extrapolate in phase space.…”

Section: Discussionmentioning

confidence: 82%

“…This confirms our general observation, that the (B)INN learns a functional form of the density in both directions, in close analogy to a fit. It also means that the uncertainty from the generative network training is not described by the simple statistical scaling we observed for simpler networks [63,64] and instead points towards a GANplification-like [4] pattern.…”

Section: Marginalizing Phase Spacementioning

confidence: 77%

“…In recent years it has been shown that modern machine learning can improve LHC event simulations in many ways [1]. Promising techniques include generative adversarial networks (GAN) [2][3][4], variational autoencoders [5,6], and normalizing flows [7][8][9][10][11], including invertible networks (INNs) [12][13][14]. They can improve phase space integration [15,16], phase space sampling [17][18][19], and amplitude computations [20,21].…”

mentioning

confidence: 99%

See 2 more Smart Citations

Understanding Event-Generation Networks via Uncertainties

Bellagente¹,

Haussmann²,

Luchmann³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Following the growing success of generative neural networks in LHC simulations, the crucial question is how to control the networks and assign uncertainties to their event output. We show how Bayesian normalizing flow or invertible networks capture uncertainties from the training and turn them into an uncertainty on the event weight. Fundamentally, the interplay between density and uncertainty estimates indicates that these networks learn functions in analogy to parameter fits rather than binned event counts. Content 4 LHC events with uncertainties 18 5 Outlook 20 References 21

show abstract

Section: Discussionmentioning

confidence: 82%

Section: Marginalizing Phase Spacementioning

confidence: 77%

mentioning

confidence: 99%

See 1 more Smart Citation

Understanding Event-Generation Networks via Uncertainties

Bellagente¹,

Haussmann²,

Luchmann³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Neural networks based data augmentation in high energy collision simulations have been addressed, for example, in Refs [53][54][55]…”

mentioning

confidence: 99%

Jets and Photons Spectroscopy of Higgs-ALP Interactions

Alves

Dias

Lopes

2021

Preprint

View full text Add to dashboard Cite

Axion-like particles (ALPs) and Higgs bosons can interact in scalar sectors beyond the Standard Model, leading the Higgs boson to decay into pairs of gluons and photons through the ALP interaction and giving rise to resonances in the decay products of the process h → aa → gg +γγ, resembling a spectral lines analysis. We explore this signature to constrain an ALP effective field theory formulation and show that the forthcoming runs of the LHC will be capable to probe the ALP-Higgs interaction in the ALP mass range from 0.5 to 60 GeV using an automatized search strategy that adapts to different ALP masses in inclusive jets plus photons final states. Such interaction can also be tested in mass regions where the two and four-photon search channels are currently ineffective.

show abstract

“…Naively, it seems that the answer must be no, based on the same reasoning that motivates the argument that a generative network cannot produce more information than exists in its statistically limited training data set. However, this argument fails to account for the implicit knowledge embed-ded in the architecture of the network, which can contribute information in the same manner as a functional fit [61]. A super-resolution network applied to LHC jets combines the information from the low-resolution image with QCD knowledge extracted from the training data, for instance the underlying theoretical principles of soft and collinear splittings combined with mass drop patterns.…”

mentioning

confidence: 99%

How to GAN Higher Jet Resolution

Baldi¹,

Blecher²,

Butter³

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

QCD-jets at the LHC are described by simple physics principles. We show how superresolution generative networks can learn the underlying structures and use them to improve the resolution of jet images. We test this approach on massless QCD-jets and on fat top-jets and find that the network reproduces their main features even without training on pure samples. In addition, we show how a slim network architecture can be constructed once we have control of the full network performance.

show abstract

GANplifying Event Samples

Cited by 8 publications

References 47 publications

Understanding Event-Generation Networks via Uncertainties

Understanding Event-Generation Networks via Uncertainties

Jets and Photons Spectroscopy of Higgs-ALP Interactions

How to GAN Higher Jet Resolution

Contact Info

Product

Resources

About