DANCE Machine Learning Workshop 2020

3 Aug 2020, 08:00 → 7 Aug 2020, 14:05 US/Pacific

https://lbnl.zoom.us/j/94735075450?pwd=a2U1RGZMODJnZ2xWWnFoc25nbDhuUT09

Christopher Tunnell (Rice), Igor Ostrovskiy, Kazuhiro Terao, Scott Kravitz (Lawrence Berkeley National Lab)

Dark-matter and Neutrino Computation Explored (DANCE) Machine Learning Workshop 2020

Slido links:

Live questions, careers panel

Recordings can be found in the timetable. In some cases, they are attached to the specific talk, i.e. "detailed" view, in other cases to the session or the "intro" or "review" at the start of the day. Unedited versions are being added first, followed by edited versions when available for the respective session being recorded.

Building on the success of the DANCE computing workshop, DANCE-ML will provide a forum to share and discuss machine learning applications in the context of the dark matter direct detection and neutrino physics community. As these experiments grow in size and complexity, the ability to extract as much useful information as possible from the underlying data becomes crucial. The goal of this workshop is to bridge the gap between broader scientific machine learning clinics and workshops geared toward large accelerator experiments, so that we can better solve the unique challenges of our shared domain. All those interested in machine learning within or adjacent to these areas are encouraged to attend, regardless of position or experience.

Due to the ongoing COVID-19 epidemic, this workshop will be held virtually via Zoom rather than in-person at LBNL as originally planned.

The registration deadline is July 12 now Aug 2. Attendees are encouraged to register as soon as convenient to facilitate session planning. Registration is free, but required for all attendees.

If you would like to share on Twitter, please use #DANCEML.

Program

The first day will be spent on interactive tutorials (one aimed at novice ML practitioners, one more advanced), with each of the four subsequent days centered around a topic identified as a priority at DANCE 2019:

1. Unsupervised/semi-supervised learning
2. Reconstruction and triggers
3. Interpretability and uncertainty estimation
4. Simulation and all other topics

In each topical session, we will have invited talks, including new ideas from outside our immediate community, as well as brief summaries of ongoing work in the field. Following these presentations will be time for roundtable discussion of current issues and how to solve them. We specifically encourage presentation and discussion of developing techniques and roadblocks experienced while exploring them - final results are not required. 

Organizing Committee

Scott Kravitz - LBNL
Igor Ostrovskiy - University of Alabama
Kazuhiro Terao - SLAC
Christopher Tunnell - Rice University

DANCE-ML_2020_Group_Photo.png

Monday, 3 August

08:00 → 10:00 Tutorial: A Practical Introduction to Machine Learning in Physics

ZoomRecording_Aug3_MachineLearningTutorial_raw.mp4

08:00 Getting Started with Machine Learning

https://github.com/swkravitz/dance-ml-2020-tutorial

Speaker: Scott Kravitz (Lawrence Berkeley National Lab)

Github repo for tutorial

ZoomRecording_Aug3_MachineLearningTutorial_Part1.mp4

ZoomRecording_Aug3_MachineLearningTutorial_Part2.mp4

ZoomRecording_Aug3_MachineLearningTutorial_raw.mp4

10:00 → 10:30 Break

10:30 → 12:00 Processing of Graph-structured Data: An Introduction

Speakers: Waheed Bajwa (Rutgers), Venkat Roy

Tutorial Slides (Processing of Graph-structured Data: An Introduction)

ZoomRecording_Aug3_GraphsTutorial_Comp.mp4

ZoomRecording_Aug3_GraphTutorial_raw.mp4

Sample Matlab Code

• circulant.m
• denoising_example.m
• evd_Laplacian.m

Tuesday, 4 August

08:00 → 08:15 Introduction to DANCE-ML and Workshop

Speakers: Christopher Tunnell (Rice), Scott Kravitz (Lawrence Berkeley National Lab)

intro.pdf

ZoomRecording_Aug4_FullWorkshop_raw.mp4

ZoomRecording_Aug4_Intro-Unlabelled-WF-Analysis_Comp.mp4

08:15 → 09:10 Unsupervised/semi-supervised learning

08:15 A readily-interpretable fully-convolutional autoencoder-like model for unlabelled waveform analysis

Speaker: Ben Krikler (Bristol)

200804 Unlabeled waveform ML.pdf

08:45 PMT waveform analysis in neutrino experiments

Speaker: Dacheng Xu

main.pdf

ZoomRecording_Aug4_PMT-WF-Analysis.mp4

09:10 → 09:20 Break

09:20 → 10:30 Unsupervised/semi-supervised learning

09:20 Estimation of Orientation and Camera Parameters from Cryo-Electron Microscopy Images with Variational Autoencoders and Generative Adversarial Networks

Speakers: FREDERIC POITEVIN (LCLS, SLAC), Nina Miolane

2020_DANCE_Miolane_Poitevin.pptx

ZoomRecording_Aug4_Cryo-EM_Comp.mp4

09:50 LUX Summary

Speaker: Scott Kravitz (Lawrence Berkeley National Lab)

LUX ML Overview DANCE-ML 2020.pdf

ZoomRecording_Aug4_LUX-Summary_Comp.mp4

10:10 Anomaly finding in the LZ dark matter experiment

Speaker: Chami Amarasinghe (University of Michigan)

Anomaly Finding in The LZ Dark Matter Experiment

ZoomRecording_Aug4_Anomaly-Finding_Comp.mp4

10:30 → 11:00 Break

11:00 → 11:30 Unsupervised/semi-supervised learning: Roundtable discussion

11:00 Room 1: Anomaly finding - applications and limitations

11:00 Room 2: How to design + when to apply generative models

11:30 → 12:00 Open discussion

Snowmass planning

Wednesday, 5 August

08:00 → 08:15 Review

08:15 → 09:15 Reconstruction and whitepaper

08:15 Holistic Machine Learning Approach to Particle-Flow Reconstruction in Imaging Neutrino Detectors

Speakers: Francois Drielsma, Francois Drielsma (SLAC)

drielsma_dance_ml_lartpc_particle_flow.pdf

ZoomRecording_Aug5_Intro-Holistic-ML_Comp.mp4

08:45 Machine Learning at the Edge: Intelligent Data Triage in Real Time

Speakers: Prof. Audrey Corbeil Therrien (Université de Sherbrooke), Ryan Coffee

Therrien_Coffee_DANCE-ML2020.pdf

ZoomRecording_Aug5_DataTriage_Comp.mp4

09:15 → 09:25 Break

09:25 → 10:30 Reconstruction and whitepaper

09:25 Signal corrections of XENON1T Dark Matter detector

Speaker: Tianyu Zhu (Columbia University)

DANCE2020.pdf

ZoomRecording_Aug5_XENON-Group_Comp.mp4

09:35 Graph Neural Network for Position Reconstruction

Speaker: Shixiao Liang (Rice University)

GNN_for_PosRec_SLiang_DANCE2020.pdf

09:45 Convolutional neural network for S2 position reconstruction

Speaker: Andrii Terliuk (MPIK, Heidelberg / Heidelberg University)

Terliuk_20200803_XENONnT_S2CNN_reco.pdf

09:55 Machine Learning at SuperCDMS

Speaker: To Chin Yu

MachineLearningCDMS.pdf

ZoomRecording_Aug5_SuperCDMS_Comp.mp4

10:15 Energy Reconstruction with Variational Autoencoders in SuperCDMS

Speaker: Ashley Li (University of British Columbia)

SuperCDMS_VAE_Slides_(FINAL).pdf

10:30 → 11:00 Break

11:00 → 11:30 Reconstruction and whitepaper: Roundtable discussion

11:00 Room 1: Snowmass planning (cross-collaboration LOI)

ZoomRecording_Aug5_Snowmass-LOI_Comp.mp4

11:00 Room 2: Position and energy reconstruction

11:30 → 12:00 Open discussion

Snowmass planning

Thursday, 6 August

08:00 → 08:15 Review

08:15 → 09:30 Interpretability and uncertainty estimation

08:15 Mapping Machine-Learned Physics into a Human-Readable Space

Speaker: Daniel Whiteson

DANCEML_interp.pdf

ZoomRecording_Aug6_Intro-ML-Physics-Human-Readable_Comp.mp4

08:45 PandaX-III, Investigation of possible features for the classification of signal and background with CNN

Speakers: Shangning Xia, Shaobo Wang, Suizhi Huang

PandaX-III_DANCE-ML.pdf

PandaX-III_DANCE-ML.pptx

ZoomRecording_Aug6_PandaX-Sig-BG-CNN_Comp.mp4

09:05 Machine learning for Neutrinoless Double Beta Decay in NEXT

Speaker: Corey Adams (Argonne National Laboratory)

DANCE-ML-NEXTpptx.pdf

ZoomRecording_Aug6_ML-0vbb-NEXT_Comp.mp4

09:30 → 09:40 Break

09:40 → 10:30 Interpretability and uncertainty estimation

09:40 EXO Summary

Speaker: Igor Ostrovskiy

EXO_summary_DANCEML_Ostrovskiy_Aug2020.pdf

ZoomRecording_Aug6_EXO-Group_Comp.mp4

10:00 Towards Deep Neural Network Analysis of Xe137 Decays to the Excited state of Cs137 in EXO-200 Data

Speaker: Seth Thibado (University of Massechusetts)

Towards DNN Analysis of Xe137 decays to Cs137 in EXO-200.pdf

10:10 Simulating scintillation signals in EXO-200 with GANs

Speaker: Shaolei Li (University of Illinois at Urbana-Champaign)

workshop-slides-Aug-shaolei.pdf

10:30 → 11:00 Break

11:00 → 11:30 Interpretability and uncertainty estimation: Roundtable discussion

11:00 Room 1: Ensuring robustness to systematics

11:00 Room 2: Quantifying uncertainties in inference

ThuDiscussionRoom2UncertaintiesInDNN02.mp4

11:30 → 12:00 Open discussion

Snowmass planning

Friday, 7 August

08:00 → 08:15 Review

08:15 → 09:20 Simulation and all other topics

08:15 Measurements using all available information with OmniFold

Speaker: Benjamin Nachman

SimulationBasedInferenceDANCE.pdf

ZoomRecording_Aug7_Intro-OmniFold_Comp.mp4

08:45 Lessons from Industry

Speaker: Manuel Weber (Descartes Labs)

DANCE ML Workshop 2020 Lessons from industry.pdf

ZoomRecording_Aug7_Ind-Lessons_Comp.mp4

09:20 → 09:30 Break

09:30 → 10:30 Simulation and all other topics

09:30 Reconstruction and machine learning for water Cherenkov detectors

Speakers: Nick Prouse (TRIUMF), Patrick de Perio

DANCE-ML_WatChMaL.pdf

09:50 Summary of Machine Learning Applications for the COHERENT Collaboration

Speaker: Peibo An (Duke University)

DANCEworkshopCOHERENT.pdf

ZoomRecording_Aug7_ML-COHERENT_Comp.mp4

10:10 Physics-informed Gaussian Process for Real-time Tuning and Control of Particle Accelerators

Speaker: Adi Hanuka (SLAC)

ZoomRecording_Aug7_Gaussian-Process-RT-Tuning_Comp.mp4

10:30 → 11:00 Break

11:00 → 12:00 Careers panel