2nd IML Machine Learning Workshop

Contribution List

31. Welcome

Lorenzo Moneta (CERN), Markus Stoye (CERN), Paul Seyfert (CERN), Rudiger Haake (CERN), Steven Randolph Schramm (Universite de Geneve (CH))

09/04/2018, 09:00

Session 1

29. DarkMachines

Tommaso Dorigo (Universita e INFN, Padova (IT))

09/04/2018, 09:20

Session 1

http://www.darkmachines.org/

45. Overview of ML in HEP

Luke Percival De Oliveira

09/04/2018, 09:35

Session 1

24. DeepJet: a deep-learned multiclass jet-tagger for slim and fat jets

Mauro Verzetti (CERN), Jan Kieseler (CERN), Markus Stoye (CERN), Huilin Qu (Univ. of California Santa Barbara (US)), Loukas Gouskos (Univ. of California Santa Barbara (US))

09/04/2018, 11:00

Session 1

We present a customized neural network architecture for both, slim and fat jet tagging. It is based on the idea to keep the concept of physics objects, like particle flow particles, as a core element of the network architecture. The deep learning algorithm works for most of the common jet classes, i.e. b, c, usd and gluon jets for slim jets and W, Z, H, QCD and top classes for fat jets. The...

28. HL-LHC tracking challenge

Jean-Roch Vlimant (California Institute of Technology (US))

09/04/2018, 11:25

Session 1

At HL-LHC, the seven-fold increase of multiplicity wrt 2018 conditions poses a severe challenge to ATLAS and CMS tracking experiments. Both experiment are revamping their tracking detector, and are optimizing their software. But are there not new algorithms developed outside HEP which could be invoked: for example MCTS, LSTM, clustering, CNN, geometric deep learning and more?
We organize on...

10. Convolutional Neural Network for Track Seed Filtering at the CMS HLT

Mr Adriano Di Florio (Universita e INFN, Bari (IT))

09/04/2018, 11:50

Session 1

Collider will constantly bring nominal luminosity increase, with the ultimate goal of reaching a peak luminosity of $5 · 10^{34} cm^{−2} s^{−1}$ for ATLAS and CMS experiments planned for the High Luminosity LHC (HL-LHC) upgrade. This rise in luminosity will directly result in an increased number of simultaneous proton collisions (pileup), up to 200, that will pose new challenges for the CMS...

38. Tutorial: TMVA

Lorenzo Moneta (CERN)

09/04/2018, 14:00

Session 2

39. Introduction to the industry session

Lorenzo Moneta (CERN), Markus Stoye (CERN), Paul Seyfert (CERN), Rudiger Haake (CERN), Steven Randolph Schramm (Universite de Geneve (CH))

09/04/2018, 15:00

Session 2

40. Multilevel Optimization for Generative Models, Games and Robotics

David Pfau (Google DeepMind)

09/04/2018, 15:10

Session 2

41. Machine Learning For Enterprises: Beyond Open Source

Jean-Francois Puget (IBM Analytics)

09/04/2018, 15:50

Session 2

invited talk from IBM analytics

42. Surrogate Models for Fun and Profit

Andrey Ustyuzhanin (Yandex School of Data Analysis (RU))

09/04/2018, 17:00

Session 2

44. Industry panel

Andrey Ustyuzhanin (Yandex School of Data Analysis (RU)), David Pfau (Google DeepMind), Jean-Francois Puget (IBM Analytics)

09/04/2018, 17:40

Session 2

32. Daily announcements

Lorenzo Moneta (CERN), Markus Stoye (CERN), Paul Seyfert (CERN), Rudiger Haake (CERN), Steven Randolph Schramm (Universite de Geneve (CH))

10/04/2018, 09:00

Session 3

30. Joint Wasserstein GAN contribution

David Josef Schmidt (Rheinisch Westfaelische Tech. Hoch. (DE)), Thorben Quast (Rheinisch Westfaelische Tech. Hoch. (DE)), Jonas Glombitza (Rheinisch-Westfaelische Tech. Hoch. (DE))

10/04/2018, 09:05

Session 3

This is a merger of three individual contributions:
- https://indico.cern.ch/event/668017/contributions/2947026/
- https://indico.cern.ch/event/668017/contributions/2947027/
- https://indico.cern.ch/event/668017/contributions/2947028/

14. Generative Models for Fast Cluster Simulations in the TPC for the ALICE Experiment

Kamil Rafal Deja (Warsaw University of Technology (PL))

10/04/2018, 10:00

Session 3

Simulating detector response for the Monte Carlo-generated
collisions is a key component of every high-energy physics experiment.
The methods used currently for this purpose provide high-fidelity re-
sults, but their precision comes at a price of high computational cost.
In this work, we present a proof-of-concept solution for simulating the
responses of detector clusters to particle...

26. Adversarial Tuning of Perturbative Parameters in Non-Differentiable Physics Simulators

Michela Paganini (Yale University (US))

10/04/2018, 11:00

Session 3

In this contribution, we present a method for tuning perturbative parameters in Monte Carlo simulation using a classifier loss in high dimensions. We use an LSTM trained on the radiation pattern inside jets to learn the parameters of the final state shower in the Pythia Monte Carlo generator. This represents a step forward compared to unidimensional distributional template-matching methods.

22. Fast calorimeter simulation in LHCb

Egor Zakharov

10/04/2018, 11:25

Session 3

Fast calorimeter simulation in LHCb

In HEP experiments CPU resources required by MC simulations are constantly growing and become a very large fraction of the total computing power (greater than 75%). At the same time the pace of performance improvements from technology is slowing down, so the only solution is a more efficient use of resources. Efforts are ongoing in the LHC experiments to...

17. A Deep Learning tool for fast simulation

Gul Rukh Khattak (University of Peshawar (PK))

10/04/2018, 11:50

Session 3

Machine Learning techniques have been used in different applications by the HEP community: in this talk, we discuss the case of detector simulation. The need for simulated events, expected in the future for LHC experiments and their High Luminosity upgrades, is increasing dramatically and requires new fast simulation solutions. We will describe an R&D activity, aimed at providing a...

33. Tutorial: Keras/TensorFlow

Stefan Wunsch (KIT - Karlsruhe Institute of Technology (DE))

10/04/2018, 14:00

Session 4

34. Interpreting Deep Neural Networks and their Predictions

Wojciech Samek (Fraunhofer HHI)

10/04/2018, 15:00

Session 4

Invited talk, http://iphome.hhi.de/samek/

5. What is the machine learning.

Bryan Ostdiek (University of Oregon)

10/04/2018, 16:30

Session 4

Applications of machine learning tools to problems of physical interest are often criticized for producing sensitivity at the expense of transparency. In this talk, I explore a procedure for identifying combinations of variables -- aided by physical intuition -- that can discriminate signal from background. Weights are introduced to smooth away the features in a given variable(s). New networks...

7. Identifying the relevant dependencies of the neural network response on characteristics of the input space

Stefan Wunsch (KIT - Karlsruhe Institute of Technology (DE))

10/04/2018, 16:55

Session 4

The use of neural networks in physics analyses poses new challenges for the estimation of systematic uncertainties. Since the key to a proper estimation of uncertainties is the precise understanding of the algorithm, novel methods for the detailed study of the trained neural network are valuable.
This talk presents an approach to identify those characteristics of the neural network inputs that...

23. Studies to mitigate difference between real data and simulation for jet tagging

Markus Stoye (CERN), Mauro Verzetti (CERN), Jan Kieseler (CERN), Arabella Martelli (Imperial College (GB)), Oliver Buchmuller (Imperial College (GB))

10/04/2018, 17:20

Session 4

The aim of the studies presented is to improve the performance of jet flavour tagging on real data while still exploiting a simulated dataset for the learning of the main classification task. In the presentation we explore “off the shelf” domain adaptation techniques as well as customised additions to them. The latter improves the calibration of the tagger, potentially leading to smaller...

18. Particle identification at LHCb: new calibration techniques and machine learning classification algorithms

Miriam Lucio Martinez (Universidade de Santiago de Compostela (ES))

10/04/2018, 17:45

Session 4

Particle identification (PID) plays a crucial role in LHCb analyses. Combining information from LHCb subdetectors allows one to distinguish between various species of long-lived charged and neutral particles. PID performance directly affects the sensitivity of most LHCb measurements. Advanced multivariate approaches are used at LHCb to obtain the best PID performance and control systematic...

35. Daily announcements

Lorenzo Moneta (CERN), Markus Stoye (CERN), Paul Seyfert (CERN), Rudiger Haake (CERN), Steven Randolph Schramm (Universite de Geneve (CH))

11/04/2018, 09:00

Session 5

16. Fisher information metrics for binary classifier evaluation and training

Andrea Valassi (CERN)

11/04/2018, 09:05

Session 5

Different evaluation metrics for binary classifiers are appropriate to different scientific domains and even to different problems within the same domain. This presentation focuses on the optimisation of event selection to minimise statistical errors in HEP parameter estimation, a problem that is best analysed in terms of the maximisation of Fisher information about the measured parameters....

25. DeepJet: a portable ML environment for HEP

Swapneel Sundeep Mehta (Dwarkadas J Sanghvi College of Engineering (IN)), Mr swapneel mehta (IT/DB Group)

11/04/2018, 09:30

Session 5

In this presentation we will detail the evolution of the DeepJet python environment. Initially envisaged to support the development of the namesake jet flavour tagger in CMS, DeepJet has grown to encompass multiple purposes within the collaboration. The presentation will describe the major features the environment sports: simple out-of-memory training with a multi-treaded approach to maximally...

13. Event Categorization using Deep Neural Networks for ttH (H→bb) with the CMS Experiment

Marcel Rieger (RWTH Aachen University (DE))

11/04/2018, 09:55

Session 5

The analysis of top-quark pair associated Higgs boson production enables a direct measurement of the top-Higgs Yukawa coupling. In ttH (H→bb) analyses, multiple event categories are commonly used in order to simultaneously constrain signal and background contributions during a fit to data. A typical approach is to categorize events according to both their jet and b-tag multiplicities. The...

27. Machine learning in jet physics

Sreedevi Narayana Varma (King's College London)

11/04/2018, 11:00

Session 5

High energy collider experiments produce several petabytes of data every year. Given the magnitude and complexity of the raw data, machine learning algorithms provide the best available platform to transform and analyse these data to obtain valuable insights to understand Standard Model and Beyond Standard Model theories. These collider experiments produce both quark and gluon initiated...

20. Recursive Neural Networks in Quark/Gluon Tagging

Taoli Cheng (University of Chinese Academy of Sciences)

11/04/2018, 11:25

Session 5

Vidyo contribution

Based on the natural tree-like structure of jet sequential clustering, the recursive neural networks (RecNNs) embed jet clustering history recursively as in natural language processing. We explore the performance of RecNN in quark/gluon discrimination. The results show that RecNNs work better than the baseline BDT by a few percent in gluon rejection at the working point of...

11. Direct Learning of Systematics-Aware Summary Statistics

Pablo De Castro Manzano (Universita e INFN, Padova (IT))

11/04/2018, 15:05

Session 6

Complex machine learning tools, such as deep neural networks and gradient boosting algorithms, are increasingly being used to construct powerful discriminative features for High Energy Physics analyses. These methods are typically trained with simulated or auxiliary data samples by optimising some classification or regression surrogate objective. The learned feature representations are then...

12. Multivariate Analysis Techniques for charm reconstruction with ALICE

Chiara Zampolli (CERN)

11/04/2018, 15:30

Session 6

ALICE is the experiment at the LHC dedicated to heavy-ion collisions. One of the key tools to investigate the strongly-interacting medium (Quark-Gluon Plasma, QGP) formed in heavy-ion collisions is the measurement of open-charm particle production. In particular, charmed baryons, such as ΛC, provide essential information for the understanding of charm thermalisation and hadronisation in the...

15. Classification of decays involving variable decay chains with convolutional architectures

Justin Tan (The University of Melbourne, Belle II)

11/04/2018, 16:30

Session 6

Vidyo contribution

We present a technique to perform classification of decays that exhibit decay chains involving a variable number of particles, which include a broad class of $B$ meson decays sensitive to new physics. The utility of such decays as a probe of the Standard Model is dependent upon accurate determination of the decay rate, which is challenged by the combinatorial background...

4. Drones: Making faster and smarter decisions with software triggers

Sean Benson (Nikhef National institute for subatomic physics (NL))

11/04/2018, 16:55

Session 6

Data collection rates in high energy physics (HEP), particularly those at the Large Hadron Collider (LHC) are a continuing challenge and require large amounts of computing power to handle. For example, at LHCb an event rate of 1 MHz is processed in a software-based trigger. The purpose of this trigger is to reduce the output data rate to manageable levels, which amounts to a reduction from 60...

37. Close-out and challenge results

Lorenzo Moneta (CERN), Markus Stoye (CERN), Paul Seyfert (CERN), Rudiger Haake (CERN), Steven Randolph Schramm (Universite de Geneve (CH))

11/04/2018, 17:20

Session 6

8. Conditional Wasserstein GANs for fast simulation of electromagnetic showers in a CMS HGCAL prototype

Thorben Quast (Rheinisch Westfaelische Tech. Hoch. (DE))

Session 3

The increased instantaneous luminosity at HL-LHC will raise the computing requirements for event reconstruction and analysis for current LHC-based experiments, hence limiting the available resources for the simulation of particles traversing matter. Developments of the performance of state-of-the-art simulation frameworks such as Geant4 are proceeding but are unlikely to fully compensate for...

19. Deep neural network-based multi-class boosted object tagger in ATLAS

Ece Akilli (Universite de Geneve (CH))

Session 5

A deep neural network-based multi-class boosted object tagger is developed in the context of a search for pair production of heavy vector-like quarks with hadronic final states in ATLAS. The four classes of the tagger are W/Z (V)-boson, Higgs-boson, top-quark and background jets. As the unambiguous identification of the origin of the jet is essential for this search, an identification...

6. Generating high-level physics variables based on Monte Carlo simulated ttH events using Wasserstein GANs

David Josef Schmidt (Rheinisch Westfaelische Tech. Hoch. (DE))

Session 3

Developing and building an analysis in high energy particle physics requires a large amount of simulated events. Simulations at the LHC are usually complex and computationally intensive due to sophisticated detector architectures. In this context, Generative Adversarial Networks (GANs) have recently caught a wide interest. GANs can learn to generate complex data distributions and produce...

43. Invited industry talk #4

Session 2

21. Machine learning in the LHCb tracking

Francesco Polci (Centre National de la Recherche Scientifique (FR)), Agnieszka Dziurda (CERN), Lucia Grillo (University of Manchester (GB)), Giulio Dujany (Universite Pierre et Marie Curie et Universite Denis Diderot ()

Session 1

The LHCb experiment at CERN operates a high precision and robust tracking system to reach its physics goals, including precise measurements of CP-violation phenomena in the heavy flavour quark sector and searches for New Physics beyond the Standard Model. Since Run2, the experiment has put in place a new trigger strategy with a real-time reconstruction, alignment and calibration, imposing...

9. Refining particle detector simulations using the Wasserstein distance in adversarial networks

Mr Jonas Glombitza

Session 3

Machine learning models, especially deep neural networks produce appropriate predictions when working on a test set similar to the training set. In physics research machine learning models are usually designed to be used for data application but trained on simulations. Therefore, differences between simulations and data can cause substantial uncertainties in the application.
Here we attempt to...

2. Research on EOS File storage Strategy Based on Access Characteristics Using Machine Learning Method

Zhenjing Cheng (INSTITUE OF HIGH ENERGY PHYSICS)

Session 1

Machine learning has been an attractive topic in high-energy physics field for many years. For example, machine learning algorithms devoted to the reconstruction of particle tracks or jets in high energy physics experiments. EOS is an open source parallel distributed file system. It has been generally used in large scale cluster computing for both physics and user use cases at IHEP, like...

46. TBA

Session 6

3. Training Deep Learning Models on Many-Core Processors

Mr David Nonso Ojika (University of Florida (US))

Session 5

Leveraging on our previous work on developing DNN-based classification models for Higss events [1], we turn to CNN-based classification models for muon events. Using Intel Knights Landing (KNL) processors, we present performance metrics on training convolutional neural networks (CNNs) on multiple KNL computing nodes for the task of muon identification (i.e "high Pt" or "low Pt"). This work is...