Deep Learning in the Natural Sciences

28 Feb 2019, 08:45 → 1 Mar 2019, 18:00 Europe/Berlin

Hamburg

A workshop to connect researchers working on Deep Learning in the natural sciences with a focus on astronomy, particle physics and photonics.

The program consists of invited speakers, submitted contributions and student talks with generous room for discussion.

The workshop will take place at:
Universität Hamburg
Jungiusstraße 9, Hörsaal I (Wolfgang Pauli-Hörsaal)
20 355 Hamburg

 

Thursday, 28 February

Pre-Workshop introduction and arrival

1 Registration & Sign-Up

2 Deep Learning Basics I

Speaker: Gregor Kasieczka (Hamburg University (DE))

DeepLearning_DLHAM_Basics.pdf

3 Deep Learning Basics II

Speaker: Gregor Kasieczka (Hamburg University (DE))

12:00 Lunch

Talk: Talks I

Convener: Marcus Brueggen

4 Opening

Speakers: Gregor Kasieczka (Hamburg University (DE)), Marcus Brueggen

DLHAM_Intro.pdf

5 Collision Course: Particle Physics as a Machine-Learning Testbed

Speaker: Jesse Thaler (MIT)

jthaler_Hamburg_Collision.pdf

6 Machine Learning Techniques in Cosmological Simulation

Speaker: Claudio Gheller

hamburg-2019-1.pdf

hamburg-2019-1.pptx

7 Low-dose X-ray Imaging with Deep Neural Networks

Speaker: Xiaogang Yang

15:00 Coffee Break

Talk: Talks II

8 Machine Learning Techniques in Astroparticle Physics

Speaker: Dominik Elsaesser

9 Rise of the Tagging Machines

Speakers: Tilman Plehn (Heidelberg University), Tilman Plehn

hamburg_19.pdf

10 Machine Learning for Diffractive Imaging and Crystallography

Speaker: Filipe Maia

Friday, 1 March

Talk: Talks III

Convener: Johannes Haller (CERN)

11 Deep Learning in Particle and Astroparticle Physics

Speaker: Martin Erdmann (Rheinisch Westfaelische Tech. Hoch. (DE))

Autonomous_Erdmann.pdf

12 A metric for collider events

When are two collider events similar? In this talk, I will answer this question by developing a metric between the radiation patterns of events. The metric is based on the well known earth mover’s distance, and intuitively is the minimum “work” required to rearrange the energy flow of one event into the other. With a metric in hand, I will discuss and demonstrate numerous tools for analyzing and visualizing the space of events for collider applications.

Speaker: Patrick Komiske (Massachusetts Institute of Technology)

PTK_DLNS_Hamburg_compressed.pdf

13 Radio Galaxy Classifications with Deep Learning

Speaker: Vesna Lukic

Hamburg_RG_deep_learning_final.pdf

14 Particle identification on the DAMPE experiment

The Dark Matter Particle Explorer (DAMPE) is a space-borne particle detector and cosmic rays observatory in operations since 2015, equipped with alongside other instruments a deep calorimeter able to detect electrons up to an energy of 10 TeV and cosmic rays up to 100 TeV. The large proton and ion background in orbit requires a powerful electron identification method. We explore a neural network based approach to an on-orbit particle identification problem. We present the issues that arise from the constraints of particle physics and our experiment, notably the difference between training set based on simulated (Monte Carlo) data, and the application set based on real unlabeled data, leading to a trade-off between performances and general usability.

Speaker: David Francois Droz (Universite de Geneve (CH))

Hamburg.pdf

Hamburg.pptx

15 Machine learning with augmentation for boosting di-Higgs searches at the LHC

Speaker: Won Sang Cho (Seoul National University)

diHiggs_DL4NS_wscho.pdf

11:00 Coffee Break

Talk: Talks IV

Convener: Gregor Kasieczka (Hamburg University (DE))

16 Application of Generative Models to Natural Science

Speakers: Fedor Ratnikov, Fedor Ratnikov (Yandex School of Data Analysis (RU))

GenerativeModels_HH_190301.pdf

17 Towards data-driven particle physics classifiers

Deep learning in particle physics often relies on imperfect simulations due to the lack of real labelled data, which risks learning mismodeling artifacts rather than the underlying physics. In this talk, I discuss the prospects for training classifiers directly on collider data using mixed samples, drawing from techniques in weak supervision and topic modeling. Using the example of quark versus gluon jet classification, I demonstrate how these ideas allow data-driven classifiers to be trained and actually provide an operational definition of the underlying categories.

Speaker: Eric Metodiev (Massachusetts Institute of Technology)

Hamburg2019_Metodiev.pdf

18 Autoencoding New Physics

Speakers: Jennifer Thompson (ITP Heidelberg), Jennifer Thompson (ITP Heidelberg)

QCDorWhat.pdf

19 CNN Classification of X-ray Selected Clusters

Speaker: Matej Kosiba

20 Closing Discussion

Speaker: Peter Schleper (Hamburg University (DE))