Simulation of Extensive Air Showers with Deep Neural Networks - - PowerPoint PPT Presentation

▶
simulation of extensive air showers with deep neural
SMART_READER_LITE
LIVE PREVIEW

Simulation of Extensive Air Showers with Deep Neural Networks - - PowerPoint PPT Presentation

Dec 24, 2022 503 likes •799 views

Simulation of Extensive Air Showers with Deep Neural Networks Marcel Kpke Auger Youngster Meeting (2019) INSTITUTE FOR NUCLEAR PHYSICS (IKP), FACULTY OF PHYSICS KARLSRUHE INSTITUTE OF TECHNOLOGY (KIT) www.kit.edu KIT The Research

slide-1
SLIDE 1

KIT – The Research University in the Helmholtz Association

INSTITUTE FOR NUCLEAR PHYSICS (IKP), FACULTY OF PHYSICS KARLSRUHE INSTITUTE OF TECHNOLOGY (KIT)

www.kit.edu

Simulation of Extensive Air Showers with Deep Neural Networks

Marcel Köpke Auger Youngster Meeting (2019)

slide-2
SLIDE 2

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 2 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

CORSIKA 7 [1]

Extensive air shower Monte Carlo simulation framework Different types of interaction models (EPOS-LHC, QGSJET, SIBYLL, ...)

1 TeV Proton 1 TeV Iron 10 TeV Proton 10 TeV Iron

slide-3
SLIDE 3

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 3 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Motivation

The time complexity of CORSIKA 7 simulations rises approximately linearly with the primary particle energy

slide-4
SLIDE 4

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 4 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Thinning

Reduces (effective) particle content by particle-aggregation Preserves shower properties to leading order Reduces shower-to-shower fluctuations

slide-5
SLIDE 5

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 5 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Why Neural Networks?

Can run on specialized hardware (GPU / TPU) Automatic parallelization (TensorFlow) Automatic reduction to essential features Training can fix meta-parameters Adjustable accuracy possible

slide-6
SLIDE 6

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 6 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Combination of linear and non-linear functions Training via loss function / metric on data pairs

Neural Networks

slide-7
SLIDE 7

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 7 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Train discriminator on real (1) and generated (0) data Train generator to outsmart the discriminator

Generative Adversarial Neural Network (GAN)

slide-8
SLIDE 8

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 8 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Train discriminator on real (1) and generated (0) data Train generator to outsmart the discriminator

Training: Discriminator (Part 1)

slide-9
SLIDE 9

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 9 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Train discriminator on real (1) and generated (0) data Train generator to outsmart the discriminator

Training: Sampling

slide-10
SLIDE 10

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 10 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Train discriminator on real (1) and generated (0) data Train generator to outsmart the discriminator

Training: Discriminator (Part 2)

slide-11
SLIDE 11

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 11 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Train discriminator on real (1) and generated (0) data Train generator to outsmart the discriminator

Training: Generator

slide-12
SLIDE 12

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 12 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Train discriminator on real (1) and generated (0) data Train generator to outsmart the discriminator

Training: Result

slide-13
SLIDE 13

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 13 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Train discriminator on real (1) and generated (0) data Train generator to outsmart the discriminator

Training: Result

slide-14
SLIDE 14

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 14 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Train discriminator on real (1) and generated (0) data Train generator to outsmart the discriminator

Training: Result

slide-15
SLIDE 15

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 15 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Train discriminator on real (1) and generated (0) data Train generator to outsmart the discriminator

Training: Result

slide-16
SLIDE 16

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 16 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

First Test (CONEX)

CONEX: Hybrid Extenisve Air Shower Simulation

–

first: Monte Carlo until energy threshold (3D)

–

then: cascade equation solver (1D)

–

provides longitudinal profile only

–

runtime: seconds – minutes Configuration:

–

E = 1E17 ... 1E19 eV

–

Zenith = 0 ... 65 deg

–

Azimuth = -180 ... 180 deg Generated ~187k datapoints

slide-17
SLIDE 17

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 17 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Shower-to-Shower Fluctuations

slide-18
SLIDE 18

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 18 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Shower-to-Shower Fluctuations

slide-19
SLIDE 19

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 19 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Shower-to-Shower Fluctuations

slide-20
SLIDE 20

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 20 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

CONEX vs. GAN

slide-21
SLIDE 21

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 21 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

CONEX vs. GAN

slide-22
SLIDE 22

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 22 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

CONEX vs. GAN

slide-23
SLIDE 23

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 23 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Shower Library

Shower library required for analyses and model training Trained model = effective compression of shower library

slide-24
SLIDE 24

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 24 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

What‘s next?

Fix it (oversampling, architecture, ...) (Meta)parameter tests Test adversarial vulnerability Template matching/reconstruction Refining with data

slide-25
SLIDE 25

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 25 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Backup

slide-26
SLIDE 26

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 26 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Fast Implicit Simulation Heuristic (FISH)

Autoencoder with Adversarial Metric Simulation Input (SI) can be extended with meta-parameters Discriminator can be refined with real measurements

slide-27
SLIDE 27

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 27 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

Adjustable Accuracy [2]

ResNet Translate to ordinary differential equation (ODE) Solve with standard ODE solver Adapt solver accuracy on the fly (training: high, inference: low)

slide-28
SLIDE 28

Institute for Nuclear Physics (IKP), Faculty of Physics Karlsruhe Institute of Technology (KIT) 28 Marcel Köpke: Simulation of Extensive Air Showers with Deep Neural Networks 23.09.2019

References

Title picture: Karlsruhe Castle - Meph666 [CC BY-SA 3.0] https://commons.wikimedia.org/wiki/File:Karlsruhe-Schloss-meph666- 2005-Apr-22.jpg Backup picture: Photo by Anthony from Pexels [1] CORSIKA 7: https://www.ikp.kit.edu/corsika/ [2] „Neural Ordinary Differential Equations“ - Ricky T. Q. Chen, Yulia Rubanova, Jesse Bettencourt, David Duvenaud – arXiv: 1806.07366