4th ATLAS Machine Learning Workshop

11 Nov 2019, 09:00 → 15 Nov 2019, 18:00 Europe/Zurich

6/2-024 - BE Auditorium Meyrin (CERN)

The forth workshop on the use of machine learning (ML) across the ATLAS Collaboration will be held at CERN November 11-15, 2019. It follows the the first, second, and third in Atlas Machine Learning workshops, and will be similar in focus.

• The agenda will be structured as follows:
  • Monday will consist of open sessions where speakers from other experiments, theorists, and from the Machine Learning community will be invited. 
  • Tuesday will be devoted to tutorials
  • Wednesday and Thursday will be dedicated to ATLAS speakers. Abstracts submitted via indict (left column) will be accepted until October 25th. 
  • Friday morning will consist of a close-out session.
  • The reminder of Friday will consist of "expert" sessions where focus will be on technical issues and planning for next following year.
• Participation is free
• Please please register (left hand column) and specify whether you'll be physically present at CERN during most of the workshop.

Contact: ATLAS ML conveners, Dan Guest and Amir Farbin.

Monday, 11 November

Public: Monday Public Session

12:00 Lunch

Public: Monday Afternoon

1 Introduction

Speakers: Amir Farbin (University of Texas at Arlington (US)), Dan Guest (University of California Irvine (US))

Intro slides

2 Dark Machines

Speaker: Bob Stienen

Dark Machines @ AML CERN.pdf

3 A deep neural network for simultaneous estimation of b quark energy and resolution for the CMS experiment

(note that speaker has to finish before 4pm)

Speaker: Nadezda Chernyavskaya (Eidgenoessische Tech. Hochschule Zuerich (CH))

N_Chernyavskaya_ML_ATLAS_11_11_2019.pdf

4 DNNLikelihood

We introduce the DNNLikelihood, a novel framework to easily encode, through Deep Neural Networks (DNN), the full experimental information contained in complicated likelihood functions (LFs). We show how to efficiently parametrize the LF, treated as a multivariate function of parameters and nuisance parameters with high dimensionality, as an interpolating function in the form of a DNN predictor. We do not use any Gaussian approximation or dimensionality reduction, such as marginalization or profiling over nuisance parameters, so that the full experimental information is retained. The procedure applies to both binned and unbinned LFs, and allows for an efficient distribution to multiple software platforms, e.g. through the framework independent ONNX model format. The distributed DNNLikelihood could be used for different use cases, such as re-sampling through Markov Chain Monte Carlo techniques, possibly with custom priors, combination with other LFs, when the correlations among parameters are known, and re-interpretation within different statistical approaches, i.e. Bayesian vs frequentist. We discuss the accuracy of our proposal and its relations with alternative approximation techniques and likelihood distribution frameworks. We apply our procedure to a pseudo experiment corresponding to a realistic LHC search for new physics already considered in the literature.

Speaker: Riccardo Torre (CERN)

Torre_2019-11-11_CERN.pdf

Torre_2019-11-11_CERN.pdf

14:40 Break

5 A deep neural network-based tagger to search for new long-lived particle states decaying to jets

Speaker: Vilius Cepaitis (Imperial College (GB))

ATLAS_ML_11Nov19.pdf

6 Fast Machine Learning Inference on FPGAs for Trigger and DAQ with hls4ml

Invited talk

Speaker: Sioni Paris Summers (CERN)

hls4ml-atlas-ml.pdf

16:10 Break

Wednesday, 13 November

13:00 Lunch

7 Dinner at La Potiniere

Restaurant: https://www.lapotinieregeneve.com/

Please register:
- https://indico.cern.ch/event/844092/registrations/
- Pay at the ATLAS Secretariat before Wednesday afternoon. Price is 84 CHF per person

Thursday, 14 November

12:00 Lunch

Friday, 15 November

12:00 Lunch

Public: Friday Afternoon

8 ML Inference in CMSSW

Speaker: Huilin Qu (Univ. of California Santa Barbara (US))

ML_Inference_CMSSW_20191115_H_Qu.pdf

9 Distributed training and optimization

Speaker: Dr Jean-Roch Vlimant (California Institute of Technology (US))

vlimant_AML-DistrL_Nov19.pdf

15:05 Break

10 Adversarial training for ttH(bb) classification

Event classification trained on Monte Carlo data can lead to a training bias towards the generator of the training sample, typically evaluated as a systematic error by comparing to an alternative generator model.
For the case of the search for a top-quark pair produced in association with a Higgs boson decaying to bottom-quark at the LHC, we demonstrate how adversarial domain adaptation can reduce such training bias.
A signal vs background classification network is extended by a discriminator so that the classification response is more uniform for alternative background generators.

Speaker: Paul Glaysher (DESY)

AdversarialClassifier_ML.pdf

11 Machine Learning for BSM

Speaker: Andrea Wulzer (CERN and EPFL)

talk_wulzer.pdf

12 GPUs in Reconstruction

Speaker: Dr Charles Leggett (Lawrence Berkeley National Lab (US))

2019.11.15_AtlasGPUUsage.pdf