Name: 2nd BMBF Big Data All Hands Meeting and 2nd Smart Data Innovation Conference
Start: 2017-10-11T08:25:00+02:00
End: 2017-10-12T18:15:00+02:00
Location: Karlsruher Institut of Technologie - Campus South

2nd BMBF Big Data All Hands Meeting and 2nd Smart Data Innovation Conference

from Wednesday, October 11, 2017 (8:25 AM) to Thursday, October 12, 2017 (6:15 PM)

Monday, October 9, 2017
Tuesday, October 10, 2017
Wednesday, October 11, 2017

8:45 AM Registration
Registration
8:45 AM - 9:45 AM
Room: Lobby
9:45 AM
9:45 AM - 10:15 AM
Room: Gaede-HS
Contributions

9:45 AM Welcome notes by Prof. Dr. Oliver Kraft

9:55 AM Introduction to BDAHM/SDIC by Prof. Dr. Michael Beigl

10:05 AM Logistics by Dr. Parinaz Ameri
10:15 AM Legal Process Modeling for Dataprotection – how to implement Privacy-by-Design - Beatrix Weber (MLE)
Legal Process Modeling for Dataprotection – how to implement Privacy-by-Design
- Beatrix Weber (MLE)
10:15 AM - 11:00 AM
Room: Gaede-HS On 25 May 2018 the General Data Protection Regulation (GDPR)1 will enter into force for the Internal Market. The GDPR will nearly substitute the European Data Protection Directive2 and the National Data Protection Codes as the German Data Protection Law (Bundesdatenschutzgesetz). According to the GDPR the controller shall, in particular using new technologies, implement a Data Protection Compli- ance Management System3 by adopting internal policies and undertaking appropriate technical and or- ganisational measures preventing the risks and possible damages related to the processing of personal data. Evaluating risks means that the protection of personal data is not an absolute right and must be consi- dered in relation to its function in society and balanced against other fundamental rights. Therefore, there is no prevalance of the Right of Informational Self-determination. The principle of proportionality leads to the application of the criteria as follows: * Likelihood and severity for rights and freedoms of natural persons posed by the processing, * State of the Art, * Cost of implementation and * Nature, scope, context and purposes of processing. Dataprotection-Compliance shall meet the requirements of prevention and evidence. According to the GDPR Privacy-by-Design means to implement the collection and use of personal data lawfully, fairly and in a transparent manner while designing products and services using personal data. The controller may classify the data according to what is adequate, relevant and limited to what is necessary in relation to the purposes and to the likelihood and severity of data protection infringements, e.g. in case of loss of data. Privacy-by-Default means the obligation to limit the collection and use of personal data to what is necessary for each specific purpose while relying on measures as anonymisation or pseudonymisa- tion. Offering such products or services may limit the product opportunities as specification for custo- mers needs. Critical success factor for the implementation of requirements of privacy-by-design is the integration into the product development process. Business Process Management Tools (BPM) are widely used to organize the companies activities and processes according to the business strategy and the product performance. Aim of the presentation is to show how BPM can be developped to a Legal Process Management (LPM) which integrates the requirements of Data Protection, e.g. Privacy-by-Design, into the business processes. Challenges are to bring law codes and standards into a process pattern, which is to analyse and disassemble them into individual parameters and to attribute them to single process steps. We will show the Legal Process Modeling of the Research Group „Law in Sustainability, Compli- ance and IT“ and the experience gained in the project sd-kama.
11:00 AM Coffee Break
Coffee Break
11:00 AM - 11:30 AM
Room: Lobby
11:30 AM Privacy-preserving data integration for Big Data - Erhard Rahm (Universität Leipzig) Eric Peukert (Universität Leipzig)
Privacy-preserving data integration for Big Data
- Erhard Rahm (Universität Leipzig)
- Eric Peukert (Universität Leipzig)
11:30 AM - 12:15 PM
Room: Gaede-HS Eine der großen Herausforderungen im Umgang mit Big Data liegt in der Auswertung (personenbezogener) Daten bei gleichzeitiger Wahrung des Datenschutzes und der Datensouveränität. Im ScaDS Dresden/Leipzig und am Institut für Informatik der Universität Leipzig werden dazu skalierbare Verfahren entwickelt und anhand von in der Praxis auftretenden Problemstellungen evaluiert. Ein wichtiger Fokus liegt auf Verfahren zum sog. Record Linkage oder Entity Matching. Dabei werden Entitäten (Personen, Produkte oder ähnliches) aus mehreren Quellen verknüpft, die dasselbe Realwelt-Objekt darstellen. Solche Prozesse basieren auf dem Vergleich von bestimmten Attributen (Quasi-Identifier) der zu verknüpfenden Datensätze, z.B. für das Matching von Publikationen werden Titel, Autorennamen und akademische Zugehörigkeit der jeweilige Publikationen miteinander verglichen. Personen-bezogenen Daten unterliegen in Deutschland und Europa jedoch strengen Datenschutzbestimmungen und dürfen gar nicht oder nur in verschlüsselter Form an eine dritte Partei weitergegeben werden. Daher ist das Matching von personen-bezogenen Daten problematisch – d.h. wenn sensitive Personendaten aus verschiedenen Organsation analysiert werden, muss garantiert werden, dass die Identität (Privacy) der zugehörigen Personen geschützt ist. Im Vortrag werden Technologien und Methoden zum Privacy Preserving Record Linkage (PPRL) vorgestellt, die beim Linking von sensitiven Daten alle wesentlichen Anforderungen (Qualität, Skalierbarkeit und Privacy) erfüllen. Eine PPRL-Methode muss dabei die folgenden Fragen beantworten: (1) Wie kann man sensitive Personendaten so anonymisieren, dass ein Rückschluss auf die ursprüngliche Daten unmöglich wird? (2) Wie kann man Daten anonymisieren und trotzdem ihre Ähnlichkeit für das Matching beibehalten? (3)Wie kann man große anonymisierte Datenmengen effizient vergleichen und dies ohne Qualitätsverlust? Die vorgestellten Ansätze werden anhand praxisnaher Anwendungsfälle im Bereich Gesundheit beschrieben z.B. das Verlinken von Patientendaten aus verschiedenen Krankenhäusern zur Identifikation von Zusammenhängen zwischen Krankheiten. Ein weiterer Anwendungsfall widmet sich Zensusdaten und den dabei auftretenden Datenschutzproblemen.
12:15 PM Technischer Datenschutz von personenbezogenen Daten - Matthias Gabel (KIT)
Technischer Datenschutz von personenbezogenen Daten
- Matthias Gabel (KIT)
12:15 PM - 12:45 PM
Room: Gaede-HS Etablierte Anonymitätsbegriffe wie k-Anonymität und Differential Privacy versprechen, dass die öffentliche Sicht auf personenbezogene Daten hinreichend verschleiert ist, um die Anonymität der betroffenen Personen zu gewährleisten. Doch für die korrekte Anonymisierung selbst müssen alle personenbezogenen Daten vollständig und zentralisiert vorliegen; die Betroffenen selbst haben keine unmittelbare Kontrolle mehr diese Daten. Während des Anonymisierungsvorgangs selbst sind die Daten also einer gewissen Gefahr ausgesetzt. Dieser Gefahr kann man auf verschiedene Arten begegnen: Einerseits bietet die moderne Kryptographie Möglichkeiten verschlüsselte Daten zu verarbeiten ohne diese dabei Entschlüsseln zu müssen. Diese Verfahren sind jedoch sehr aufwändig und noch nicht praktikabel. Andererseits bieten moderne Intel-Prozessoren mit der SGX-Erweiterung eine Möglichkeit, Programmcode in einer geschützten Umgebung (einer sogenannten Enklave) auszuführen. Berechnungen innerhalb dieser Enklaven sind zwar effizient durchführbar; das Sicherheitsmodell basiert aber auf der grundlegenden Annahme, dass Intel als vertrauenswürdige dritte Partei agiert. In diesem Vortrag wird aktuelle Forschung dazu vorgestellt, wie sich diese beiden Ansätze miteinander kombinieren lassen, sodass die Schwächen der beiden Ansätze abgemildert werden.
12:45 PM Lunch
Lunch
12:45 PM - 2:00 PM
Room: Lobby
2:00 PM
2:00 PM - 3:30 PM
Room: 0.014
Contributions

2:00 PM High Quality Graph and Hypergraph Partitioning - Sebastian Schlag (ITI, KIT)

2:30 PM Skalierbare graphbasierte Analyse und Business Intelligence - Erhard Rahm (Universität Leipzig) Eric Peukert (Universität Leipzig)

3:00 PM Streaming Analytics with Software AG Apama in connection with Kafka - Martin Skorsky (Software AG)
2:00 PM - 3:30 PM
Room: -1.025
Contributions

2:00 PM Research and Innovation in the Berlin Big Data Center - Jonas Traub (Technische Universität Berlin)

2:30 PM Big-Data-Kompetenzzentrum ScaDS Dresden/Leipzig – Services für Big Data - Rene Jaekel (TU Dresden)

3:00 PM Architectural designs for data analytics at HPC - Rene Jaekel (TU Dresden) Wolfgang. E. Nagel (TU Dresden / ZIH)
2:00 PM - 3:30 PM
Room: -1.014
Contributions

2:00 PM FEE- Methoden und Prozesse zur Entscheidungsunterstützung in der chemischen Industrie - David Arnu (RapidMiner GmbH)

2:30 PM AnalyzeGenomes.com: A Federated In-Memory Database Platform for Digital Health - Matthieu-P. Schapranow (Hasso Plattner Institute)

3:00 PM Das Web als Firmendatenbank: Big Data für Marketing, Vertrieb und Pressearbeit - Matthias Keller (Echobot) Nico Schlitter (KIT/SCC)
3:30 PM Coffee Break
Coffee Break
3:30 PM - 4:00 PM
Room: Lobby
4:00 PM
4:00 PM - 5:30 PM
Room: 0.014
Contributions

4:00 PM Efficiently Handling Streams from Millions of Sensors - Jonas Traub (Technische Universität Berlin)

4:30 PM AL in action: unified relational- and graph processing - Johannes Luong (Technische Universität Dresden)

5:00 PM Anomaly detection in time series data - Jan Frenzel (TU Dresden)
4:00 PM - 5:30 PM
Room: -1.025
Contributions

4:00 PM Konzept einer Big Data Infrastruktur für zukünftige Leitstellensysteme - Karl-Uwe Stucky (KIT)

4:30 PM The Smart Data Innovation Lab Research Platform - Jan Erik Sundermann (Karlsruhe Institute of Technology)

5:00 PM Thrill: High-Performance Algorithmic Distributed Batch Data Processing with C++ - Timo Bingmann (KIT)
4:00 PM - 5:30 PM
Room: -1.014
Contributions

4:00 PM BigGIS - Stream Enrichment With Temperature Data - Julian Bruns (FZI Forschungszentrum Informatik)

4:30 PM BigGIS - Einsatz von Flugrobotern zur Entscheidungsunterstützung bei Bränden - Bodo Bernsdorf (EFTAS Fernerkundung Technologietransfer GmbH)

5:00 PM Combining software heritage with Open Source, Cloud with On-Premise - Plamen Kiradjiev (IBM)
5:45 PM in front of Building 20.30
in front of Building 20.30
5:45 PM - 5:50 PM

6:30 PM ZKM Museum
ZKM Museum
6:30 PM - 7:30 PM

7:30 PM from ZKM
from ZKM
7:30 PM - 7:40 PM

8:00 PM

8:00 PM - 11:00 PM
Thursday, October 12, 2017

9:00 AM

9:00 AM - 11:30 AM
Room: 0.014

9:00 AM - 11:30 AM
Room: -1.014

9:00 AM - 11:30 AM
Room: -1.025

11:30 AM Coffee Break
Coffee Break
11:30 AM - 11:45 AM
Room: Lobby
11:45 AM SDSC-BW - Smart Data for SMEs - Andreas Wierse (SICOS BW GmbH)
SDSC-BW - Smart Data for SMEs
- Andreas Wierse (SICOS BW GmbH)
11:45 AM - 12:15 PM
Room: Gaede-HS The Smart Data Solution Center Baden-Württemberg (SDSC-BW) has been established to support small and mediuem sized enterprises (SMEs) on their way to an efficient usage os smart data analytics technology. SICOS BW and its partner, the Karlsruhe Institute for Technology (KIT) guide SMEs through the first steps into smart data analytics. The means of choice is a so-called potential analysis, where the researchers from KIT work together with the IT and application experts of the SME to create a first understanding of all issues around the application of Smart Data technology, including the availability of data, its formats, the employees involved and a first analytics session. We will present how this process is organized and how it is applied to real world scenarios. A few success stories that came out of SDSC-BW will be shown.
12:15 PM Realizing smart data by automating tabular search, integration and extraction methods - Edwin Yaqub (RapidMiner GmbH)
Realizing smart data by automating tabular search, integration and extraction methods
- Edwin Yaqub (RapidMiner GmbH)
12:15 PM - 12:45 PM
Room: Gaede-HS There is an increased focus on deriving business value from data. To exploit data, it is usually necessary to fetch it out of various silos to get a clear and holistic picture for the domain of discourse. The challenges lie in merging diverse data sets so data mining tasks can be performed on enriched data sets, which better represent the specific use case at hand. In current art, there is a lack of software tools and methods that assist in curating data from heterogenous sources to realize smarter transformations compared to their archaic forms. The problem is complicated because data exists in various formats and lies on public or private sources. These inadequacies block progress in extracting value out of data. To resolve these shortcomings, we are researching and developing user interface based tools and scientific methods in the ongoing research project “Data Search for Data Mining (DS4DM)” [1]. More concretely, we use structured data tables as basis to address *data extraction* and *data enrichment* by developing extensions for the open-source data science platform “RapidMiner”. Thus, the presented work leads to smart exploitations of data in graphically designed and highly reusable data mining processes. Data Extraction: Our work towards data extraction enables data scientists to conveniently retrieve data tables from popular sources directly into RapidMiner. These include Wikipedia articles, websites, PDF documents, online (Google) spreadsheets, etc. This adds to the out-of-box features already available in RapidMiner, which allow for reading spreadsheet documents, CSV and XML files, RDBMS databases, Cloud storage (Amazon S3) and NoSQL data-stores such as Cassandra and MongoDB, etc. Data Enrichment: Our work towards data enrichment implements the Search-Join [2] method within the graphical user interface of RapidMiner. Search-Join is a two-folds structured (tabular) search method. First, relevant data tables are searched from potentially large tabular corpus for a provided query. The tabular corpus serves as a reliable data store. Our current prototype consists of half a million data tables extracted from Wikipedia, but data can be ingested from organizational data-stores as well. The search query is comprised of an existing data table and an additional attribute, which needs to be discovered. This way, new tabular columns can be discovered and appended with imprecise or vague knowledge such as text keywords. The query is resolved by discovery algorithms that compute schema (table’s header) level and instance (table’s row) level matches for the query. This returns a space of candidate tables, which have strong contextual resemblances with the query and hence may add value to the original data table. As data search is susceptible to noise, the potentially large number of discovered tables need to be refined. This introduces data integration challenges, namely i) manual integration and ii) automatic integration of data, without which the practicality of discovered results remains of lesser value. - *Manual Data Integration:* Ideally, the data scientist needs to manually examine results and remove noisy tables, so that only value-contributing tables are considered. To guide the human in removing noise but preventing loss of informative tables, exploratory visualization techniques are developed, e.g., i) A Self-Organizing Document Map reveals how tables cluster based on similarity measures and ii) Graphical controls to manipulate intermediate outcomes of Search-Join process in real time i.e. removing noisy tables or observing distributions of certain statistical metrics among discovered candidate tables, which help to understand resemblances. - *Automatic Data Integration:* This option allows to execute the Search-Join as a fully automated process using default options, so extraction and enrichment can be operationalized. Results ------- The presented work implements a domain-independent solution to realizing smart data (through extraction and enrichment). The quality of search results considers statistical metrics such as coverage, trust, ratio and empty values, which are useful for data integration. As a result, the RapidMiner platform has been extended to incorporate data discovery and integration methods in data mining processes. Four extensions are developed and made publicly available at [3]. Acknowledgements ---------------- This work is sponsored by the German ministry of education and research (BMBF) under grant agreement number 01IS15027A-B. [1] DS4DM project website, weblink: http://ds4dm.de [2] The Mannheim Search Join Engine, C. Bizer et al., Web Semantics: Science, Services and Agents on the World Wide Web. Vol.35, Part 3, Dec. 2015. [3] The RapidMiner Marketplace, weblink: https://marketplace.rapidminer.com/UpdateServer/faces/index.xhtml
12:45 PM Lunch Break
Lunch Break
12:45 PM - 2:00 PM
Room: Lobby
2:00 PM Best Practices and Tools for Analysis Support based on SDIL Project Examples - Till Riedel (TECO KIT)
Best Practices and Tools for Analysis Support based on SDIL Project Examples
- Till Riedel (TECO KIT)
2:00 PM - 2:30 PM
Room: Gaede-HS In this talk, we discuss typical challenges based on real world examples from SDIL projects. We often find that analysis projects are comparable to a "creative chaos" and their organization is extremely depended on the preferences of the individual data scientist. We show that it is mandatory to adhere to specific requirements and design choices as projects grow larger and analytics get more complex. To support a more structured analysis process we developed a prototypic tool based on Jupyer Notebooks which we will demonstrate.
2:30 PM Establishing data marketplaces based on semantic vocabularies - Christian Mader (Frauenhofer IAIS)
Establishing data marketplaces based on semantic vocabularies
- Christian Mader (Frauenhofer IAIS)
2:30 PM - 3:00 PM
Room: Gaede-HS The role and economic value of data is increasingly recognized across various businesses and domains. Recently, capabilities for storing, processing and retrieving large amounts of data that come from various stages of industrial production workflows and product lifecyles have improved tremendously. To gain additional business value out of this data, smart data-driven services and novel value creation chains need to be established, providing a means for trustful data exchange that ensures traceability as well as the data owner’s privacy and > sovereignty over data. In addition, data consumers need a means to effectively find data that fulfills their information need and which can be used according to acceptable conditions. We argue that for establishing novel methods for creating value out of data, a common understanding of data and metadata across multiple domains is crucial. The approach we propose focuses on a shared information model based on semantic technologies that can describe and express data assets and infrastructure components in a unified way and that builds on existing models and standards. It acts as a point of reference for various stakeholders (e.g., data providers, data consumers, service providers) to publish or consume their contributions and offerings in an domain- and technology-agnostic way, building neutral and decentralized data marketplaces.
3:00 PM Coffee Break
Coffee Break
3:00 PM - 3:15 PM
3:15 PM Einführung in Maschinelles Lernen zur Datenanalyse
Einführung in Maschinelles Lernen zur Datenanalyse
3:15 PM - 6:00 PM
Room: Gaede-HS
Contributions

3:15 PM Einführung in Maschinelles Lernen zur Datenanalyse - Morris Riedel (Forschungszentrum Juelich)
Machine Data Insight Tutorial based on a Real Scenario from the SmartFactoryKL Project
Machine Data Insight Tutorial based on a Real Scenario from the SmartFactoryKL Project
3:15 PM - 6:00 PM
Room: 0.014
Contributions

3:15 PM Machine Data Insight Tutorial based on a Real Scenario from the SmartFactoryKL Project - Benedikt Krüger (IBM) Michael Vössing (KSRI)