Hasso-Plattner-Institut
Hasso-Plattner-Institut
  
Login
  • de
 

30.08.2017

Summer School on Efficient Architectures for Data Science (EADS 2017)

The Summer School on Efficient Architectures for Data Science will bring together students, researchers and industry practitioners to explore opportunities and challenges of using modern hardware resources for emerging data science applications.

When, Where, Who & What?

Hasso Plattner Institute
Potsdam | September 18 - 22, 2017

1st Summer School of the HPI Future SOC Lab

Data science applications require an adept handling of underlying hardware and software resources to achieve the required scalability and efficiency. To unleash the full potential of data science technologies, the developers' understanding needs to go beyond the mere usage of third-party library interfaces. Emerging data science applications can only benefit from improved hardware acceleration, if compute and storage resources are managed efficiently. Non-uniform memory hierarchies and heterogeneous computing architectures offer enormous opportunities but also pose new challenges for deep learning and other machine learning approaches. The Summer School on Efficient Architectures for Data Science will bring together students, researchers and industry practitioners to explore these opportunities and challenges.

The event focuses on hands-on experience and mutual exchange. It features two parts:

  • There will be expert-taught classes on data science technologies, efficient resource utilization, and advanced optimization approaches.
  • The majority of the time will be allocated to a practical software development and optimization contest, which will challenge participants to apply their knowledge in a real-world data science scenario.

We are looking forward to an exciting week of inspiration, learning, and hacking.

Prerequisites

Exceptional Master students and Ph.D. candidates with research interests in data science, machine learning, parallel computing, resource optimization, or heterogeneous computing are encouraged to apply. Programming skills in C/C++ and Python are desirable.

The target audience comprises of master students and Ph.D. students who have some experiences with data science – mainly toy examples – and little experiences with real-world challenges related to performance and scale.

Important Dates

July 28, 2017>End of Early Registration Period
September 10, 2017>Final Registration Deadline
September 18-22, 2017 > Future SOC Lab Summer School

List of Topics

  • Performance of data science techniques and algorithms
  • Scalability and efficiency of deep learning
  • Efficient resource utilization and monitoring
  • Optimization for massively parallel hardware
  • NUMA-aware programming
  • Heterogeneous and accelerated computing

Agenda

Preliminary, Subject to change.

Monday, September 18, 2017

HS 2 (Talks), H-E.11/13 (Hands-on)

09:00 – 09:30Registration
09:30 – 09:45Opening of the Summer School on Efficient Architectures for Data Science
Prof. Dr. Andreas Polze, Hasso Plattner Institute, Potsdam
09:45 – 10:30Keynote
Prof. Dr. Felix Naumann, Hasso Plattner Institute, Potsdam
Big Data Science
10:30 – 10:45Coffee Break
10:45 – 12:30Current Applied Research
Christian Bartz, Hasso Plattner Institute, Potsdam
End-to-End Detection and Recognition Systems for Computer Vision
Tim Repke, Hasso Plattner Institute, Potsdam
Analysing Business Communication
12:30 – 14:00Lunch Break
14:00 – 14:45EADS Challenge
Bernhard Rabe, HPI Future SOC Lab, Potsdam
HPI Future SOC Lab Insights
Frank Feinbube, Chairman Projects, SAP Innovation Center, Potsdam
EADS 2017 Challenge Details
14:45 – 17:00Hands-on: Challenge, Frameworks, Tools
Sven Köhler, Lena Feinbube, Carsten Walther, Max Plauth
Getting started: Infrastructure, Framework, MPI, OpenMP, CUDA, SIMD
17:00 – 17:15Wrap-up & Feedback
18:00Social Event
Barbecue at HPI

Tuesday, September 19, 2017

HS 2 (Talks), H-E.11/13 (Hands-on)

09:00 – 11:15Hands-on: Resource Management and Performance Optimization
Thomas Jakobs, Chemnitz University of Technology, Chemnitz
Optimizations for the Future: Energy and Power Efficiency
Sven Köhler, Hasso Plattner Institute, Potsdam
Parallel Programming with MPI
Felix Eberhardt, Hasso Plattner Institute, Potsdam
Optimization Guidelines for NUMA Architectures
11:15 – 11:30Coffee Break
11:30 – 12:30Hands-on: Resource Management and Performance Optimization
Sven Köhler, Hasso Plattner Institute, Potsdam
Data-parallel Programming with Vector Instructions
Max Plauth, Hasso Plattner Institute, Potsdam
An Introduction to GPU Accelerated Computing
12:30 – 14:00Lunch Break
14:00 – 17:00Hands-on: Challenge Details & Project Time
Max Plauth, Felix Eberhardt, Frank Feinbube, Sven Köhler, Carsten Walther
17:00 – 17:15Wrap-up & Feedback

Wednesday, September 20, 2017

HS 2 (Talks), H-E.11/13 (Hands-on)

09:00 – 10:00Fast Abstracts (EADS2017 Attendees)
Laura Morgenstern
Towards a NUMA-aware Task-based Fast Multipole Method
Kate Hofmann
A Framework for Evaluation of Computational Cognition
Cristian Camilo Castellanos
BDA ArchOps: An Executable Architectural Model for Big Data Analytics
Sedigheh Eslami
Privacy for Data Science: The Trade-off Between Privacy and Utility
Paul Opuchlich
Data Science for Intrusion Detection Systems
10:00 – 10:45Stories from the Industry
Dr. Rainer Kerth, Chief Architect, SAP Leonardo ML foundation
SAP Leonardo ML Foundation: Overview of SAP’s Platform for Machine Learning
10:45 – 11:00Coffee Break
11:00 – 12:30Stories from the Industry
Megha Agarwal, Data Scientist II, Pivotal Software, Inc
Agile Data Science
Dr. Fritz Schinkel, Head of Competence Center Big Data, Product Business, Fujitsu
Machine Learning for Big Data Analysis
12:30 – 14:00Lunch Break
14:00 – 17:00Hands-on: Project Time
Max Plauth, Felix Eberhardt, Frank Feinbube, Sven Köhler, Carsten Walther
17:00 – 17:15Wrap-up & Feedback
17:20Social Event
Barbecue at Innovation Center Potsdam

Thursday, September 21, 2017

HS 2 (Talks), H-E.11/13 (Hands-on)

09:00 – 12:30Hands-on: Project Time
Max Plauth, Felix Eberhardt, Frank Feinbube, Sven Köhler, Carsten Walther
12:30 – 14:00Lunch Break
14:00 – 17:00Hands-on: Project Time
Max Plauth, Felix Eberhardt, Frank Feinbube, Sven Köhler, Carsten Walther
17:00 – 17:15Wrap-up & Feedback
17:15 – 21:45Project Time
22:00Deadline Contest Submission

Friday, September 22, 2017

HS 2 (Talks), H-E.11/13 (Hands-on)

09:00 – 10:00Presentation Preparation
EADS2017 Attendees
10:00 – 12:00Project Presentations
EADS2017 Attendees
12:00 – 13:00Lunch & Jury Meeting
13:00 – 13:30Reception
13:30 – 14:00Award Ceremony & Closing Remarks
Prof. Dr. Andreas Polze, Hasso Plattner Institute, Potsdam

Prize

The best implementations from the programming contest will be awarded special prizes backed by our sponsors. Contestants will be evaluated regarding the efficiency, resource utilization, performance and exactness of their solutions.

All participants of EADS will receive a certificate for their accomplishments. The jury will comprise the organizers and sponsors.

Venue

The event will take place at Hasso Plattner Institute, Potsdam, with direct access to the HPI Future SOC Lab.

The Future SOC Lab, a cooperation of HPI and the industrial partners DellEMC, Fujitsu, SAP and Hewlett Packard Enterprise (HPE), offers an infrastructure of powerful hardware and software, including a 1000 core cluster, multi-core servers, a variety of accelerators and coprocessors, and cutting edge storage and interconnect technologies.

The Future SOC lab enables the computational acceleration of various research projects. For EADS, the lab provides a unique location, facilitating programming experiments on state-of-the-art and next-generation infrastructure. For the contest, participants will be granted exclusive access to all the Future SOC lab resources. Registration fees include the lectures with accompanying material, exercises, access to the infrastructure, all social events, and refreshments.

Accommodation and travel will have to be covered by the participants.