1. Dark Matter Research Cluster based on Computational Science
2017 KIAS-IBS Joint Workshop on Particle Physics and Cosmology
High1 resort, Jeongseon, Korea
Dark Matter Research Cluster based on Computational Science
~ 10
Kihyeon Cho(KISTI)

2. Contents Introduction What we needs for Dark Matter Research
Dark Matter Research Big Data Platform
Summary

3. Introduction

4. Project Title Dark Matter Research Cluster Leader Kihyeon Cho
Host Institute
KISTI
Partner Institutes
KASI and 15 Institutes
Fund
50,000,000 won/year * 2years
Period ~
Sponsor
NST (National Council of Science and Technology)

5. Universities (Students)
Dark Matter Research
From Collider to Astronomy
Dark Matter
Components
Standard Model
Evolution of Universe
KASI ⇒ Astronomical Data
KISTI (GSDC) ⇒ Collider Data
Origin of Matter
Standard Model
Universities (Students)

6. Goal
To make synergies for searching dark matter in experiments and theories
To make a national project in the field of dark matter
Prof. K.Y. Choi

7. Meetings Seminar and Meeting: once per month
Dark Matter Research Cluster Kickoff workshop
KISTI, October 27, 2015, 2:00-6:00 PM
2016 Dark Matter Cluster Workshop
KISTI, April 22-23, 2016
Yangyang, July 20~21, 2016
2016 Joint workshop on France-Korea
and France-Japan Particle Physics Workshop
KIAS, May 18~20, 2016
~100 persons from France, Japan, China, Korea

8. Outreach Home pages Advertisement (panel, conference)
⇒ Dark Matter Portal
Advertisement (panel, conference)
Dark Matter Research White Paper
SCOPUS Journal(KPS) Special Edition

9. What we need for Dark Matter Research

10. Scientific discovery drivers
Data
Sensors, Instruments, DB, Internet, Storage
Computational Science
Theory-Experiment-Simulation
Machine Learning
AI, Statistics, Data Mining, Algorithms, Deep Learning

11. Dark Matter Search Data Direct Collider Indirect Prof. J.C. Park
KISTI (GSDC)⇒Collider Data
KASI ⇒ Astronomical Data
Prof. J.C. Park

12. Astronomical Data
Up to ~500 PB

13. Collider Data

14. Prof. Y. S. Song

15. Theory-Experiment-Simulation => computational Science

16. From Theory Rapid update & sharing of information
Key point summary DB site by specialists
Regular meeting, Network Feedback between experimentalists and theorists
⇒ Through a (new) center
Rapid & Easy comparing check of theoretical model
Development of new numerical package
Computing power and Person power ⇒ new jobs
Cross check with astrophysics (theoretical model)
N- body simulation
Discussion with Dark Energy survey group
Prof. J.C. Park

17. From Experiment Person Power: Networking
Dedicated team/scientists for DM search for HW/SW
Networking
Networking for both Korea experimentalists in local and international institutions in DM search experiments (in Indirect, Direct and Colliders)
Regular workshop including international workshop
Enhancement communication between experimental and theoretical community
⇒ brain-storming to new ideas
Computing resources for big data :
HPC (High-Performance Computing) for MC
Storage for experimental data and theoretical model
Prof. H.D. Yoo & Y. Kwon

18. From simulations
To comply with different architecture (GPU, MIC and etc.)
To draw community interests for collateral effort

19. Dark Matter Research Big Data Platform
- Creative Alliance Project(CAP)

20. Research on Dark Matter
Issue
To combine astronomical data and HEP data based on Named Data Networking(NDN) system to make processing/analysis more efficient
Solution
Based on deep learning, we process and analyze convergence data using big data platform
R&D for Intelligence Information to open outputs
⇒Dark Matter Research Big Data Platform

21. Named Data Networking(NDN) System
GSDC
Data Production
Accelerator
(CERN,KEK Fermilab)
Astronomy
(DESI,LSST)
KASI
Data Center
Supercomputer
Theory
Intelligence
Information
(Efficiency)
Dark Matter Research Big Data Platform
Researcher
Deep Learning
Research
Data Processing
(timeliness)
KISTI
Institute
Experiment/
Observatory
NDN
Router 4
NDN router1
NDN router 2
NDN router 3
IP-less
NDN system
Named Data Networking(NDN) System
Computational Science
(reliability)
Deep
Learning
Simulation Data

22. Current Status of Data

23. Requirement of Data
E. Dobson

24. From theory-driven approach to data-driven approach
=> Named Data Networking system
E. Dobson

25. Deep Learning VS Theory Experiment Dark Matter Physics Model Data
Physics Simulation
(MadGraph/MadDM)
Compression
Event Simulation
(PHYTHIA)
Reconstruction
Simulation
Detector Simulation
(Delph/PGS/Geant4)
Analysis Tools
(PAW/ROOT)
Result

26. Deep Learning in HEP (Current)
Higgs Boson Detection (Nature Comm. 2014)
Higgs Decay (NIPS 2015, PRL 2015
Dark Knowledge, Dark Matter (JMLR C&P 2015)
Tracking in Run 4
Imaging Processing

27. T. Golling

28. Dark Knowledge Problem In HEP
Very deep and complex modes are computationally expensive.
Bad for time/CPU sensitive applications
In HEP
Possible integration of Kaggle Challenge winners
Reproduce output of massive ensemble with simple model
Trigger, real time tracking, etc
Can push models directly to hardware

30. T. Golling

32. Example

33. Summary

34. Summary(1/2) Proposing Dark Matter Research Big Data Platform
Proposed Organization
Dark Matter Research
Big Data Platform
Collaboration
IBS-CAPP
IBS-CUP
IBS-CTPU
Domestic
International
KIAS, GNUE, SNUST, SSU , Yonsei U , CNU, CAU, etc.
KNU , GSNU, Korea U, SNU, Yonsei U, JBNU CAU, CNU, etc.
Accelerator Big Data (GDDC)
Simulation Big Data
Astronomical
Big Data
Fermilab
CERN
KEK
DESI, LSST
Big Data Provider
Big Data User
Dark Matter Theory
KASI
KISTI
Dark Matter
Experiment
NDN system

35. To prove mystery of Dark Matter
Summary(2/2)
To prove mystery of Dark Matter
Astronomical Data
Theory
Accelerator Data
Dark Matter Research
Big Data Platform

36. Thank you.

37. Core Value Core Value Requirement Status Current Status Convergence SW
･Based on deep learning, convergence data processing and analysis
･Data transfer using NDN
･ Separated data processing
･ IP based data transfer
Convergence
Research
･Theory-Experiment-Simulation using convergence data
･ Theory, Experiment, Simulation respectively using separated data
Efficiency
･ Big data platform/
･ NDN based data transfer
(Time and cost reduction)
･ Each computer