Faisal Z. Qureshi | 13.01.2016 | 10:00 Uhr | Seminarraum Lakeside Labs B4.1.114

Abstract:

We developed a new method for extracting 3D flight trajectories of droplets using high-speed stereo capture. We noticed that traditional multi-camera tracking techniques fare poorly on our problem, in part due to the fact that all droplets have very similar shapes, sizes and appearances. Our method uses local motion models to track individual droplets in each frame. 2D tracks are used to learn a global, non-linear motion model, which in turn can be used to estimate the 3D locations of individual droplets even when these are not visible in any camera. We have evaluated the proposed method on both synthetic and real data and our method is able to reconstruct 3D flight trajectories of hundreds of droplets. The proposed technique solves for both the 3D trajectory of a droplet and its motion model concomitantly, and we have found it to be superior to 3D reconstruction via triangulation. Furthermore, the learned global motion model allows us to relax the simultaneity assumptions of stereo camera systems. Our results suggest that, even when full stereo information is available, our unsynchronized reconstruction using the global motion model can significantly improve the 3D estimation accuracy.

Bio:

Faisal Qureshi is an Associate Professor of Computer Science at the University of Ontario Institute of Technology (UOIT), Oshawa, Canada. He obtained a PhD in Computer Science from the University of Toronto in 2007.  He also holds an M.Sc. in Computer Science from the University of Toronto, and an M.Sc. in Electronics from Quaid-e-Azam University, Pakistan.  Prior to joining UOIT, he worked as a Software Developer at Autodesk. His research interests include sensor networks, computer vision, and computer graphics. He has also published papers in space robotics.  He has interned at ATR Labs (Kyoto, Japan), AT&T Research Labs (Red Bank, NJ, USA), and MDA Space Missions (Brampton, ON, Canada).  He is a member of the IEEE and the ACM.

Alexander Beck | 10.12.2015 | 15:00 Uhr | HS 4

Kurzfassung

IT-Systeme sind stark zusammenhängende komplexe Strukturen, so dass eine Fokussierung auf das IT-System im Ganzen nur eine bedingt effiziente Sicherheitsbewertung gewährleistet. Die Sicherheit dieser komplexen Systemlandschaften stets aktualisiert unter Berücksichtigung aller Wechselbeziehungen zwischen Komponenten dieser Systemlandschaften darzustellen, gestaltet sich mangels geeigneter Modelle schwierig. Neben diesen Modellen bilden Schwachstellenbewertungen die Grundlage für die Bewertung der Sicherheit und werden durch aufwändige Betrachtungen interpretiert, um eine Gesamtbewertung zu ermitteln. Will man dieses Vorgehen automatisieren, muss zunächst eine gemeinsame Repräsentation von Schwachstellenbewertungen vereinbart werden. Das dazu geeignete Common Vulnerability Scoring System (CVSS) ermöglicht die Bewertung einzelner Schwachstellen hinsichtlich verschiedenster Fragestellungen. Um eine Darstellung der Gesamtsicherheit zu erreichen, müssen diese Schwachstellen aggregiert werden. Unter einer Aggregation ist dabei die gemeinsame Interpretation aller im Fokus stehenden Schwachstellen zur Erreichung einer Gesamtbewertung zu verstehen, welche auf Basis eines neuronalen Netzwerkes erfolgt. Das neuronale Netz ist ein lernfähiges Konzept der Informatik, mit dem es möglich ist auf Basis definierter Eingabeparameter ein definiertes Ergebnis zu modellieren. Dabei wird das Netz trainiert und die Ergebnisse mit Ergebnissen der bisherigen manuellen Bewertungsverfahren verglichen, bis eine entsprechende Qualität der automatisch ermittelten Ergebnisse gesichert ist.

Alexander Beck ist seit 2011 bei der Volkswagen AG tätig. Zuvor studierte er Informatik an Hochschule Harz und Otto-von-Guericke-Universität Magdeburg unter anderem mit den Schwerpunkten Datenintensive Systeme und Sicherheit. Im Rahmen seiner Dissertation erforscht er Verfahren zur automatisierten Sicherheitsbewertung von komplexen heterogenen IT-Infrastrukturen auf Basis neuronaler Netze.
Beruflich war er mehrere Jahre in der Informationssicherheit im Volkswagen Konzern tätig und beschäftigte sich mit den Themen Authentifizierung und Verschlüsselung. Aktuell arbeitet er im Bereich IT Projekt- und Programmmanagement der Volkwagen Financial Services AG und verantwortet IT Projekte im In- und Ausland.

Dr. Christian Beecks | 24.11.2015 | 16:00 Uhr | E.2.42

Abstract

With the advent of social networks and the advancement of powerful internet-enabled mobile devices, millions of users are able to easily generate, process, and share multimedia data at billion-scale every single day. The resulting multitude and versatility of multimedia data made available in the Internet challenge todays’ data management and analysis algorithms. In many research and application areas including information retrieval, data mining, and computer vision, users are no longer satisfied with keyword-based access but want to search, browse, explore, and analyze multimedia data according to content-based characteristics. One fundamental operation underlying many data analysis algorithms is similarity search which aims at retrieving the most similar multimedia objects with respect to a query. In order to carry out similarity search for query-like multimedia objects, the way of modeling similarity is of major significance due to its impact on efficiency and effectiveness.

In this talk, I will present my ongoing research in this fascinating field and highlight future research directions. More specifically, I will show how to approach similarity between multimedia data objects by means of gradient-based signatures in order to facilitate data analysis with high efficiency and efficacy.

Christian Beecks is a postdoctoral researcher in the data management and data exploration group at RWTH Aachen University, Germany. His research interests include efficient and adaptive multimedia data analysis, distance-based multimedia indexing and query processing, and real-time data management.

Prof. Fabio Stella | 11.11.2015 | 10:00 Uhr | E.1.42

Abstract

Non stationary continuous time Bayesian networks are presented and described. They allow to model systems where conditional independence relationships are allowed to change over time at discrete points in time. They build on the main blocks of continuous time Bayesian networks and non stationary dynamic Bayesian networks. The seminar presents the problem of non-stationary structural learning for such probabilistic graphical models and describes solution algorithms for three different settings. Furthermore, we present preliminary results of non stationary structural learning of Continuous Time Bayesian Networks on the following biological datasets; drosophila saccharomyces cerevisiae and songbird.

Fabio Stella is an associate professor at the Dipartimento di Informatica, Sistemistica e Comunicazione of the Università degli Studi di Milano-Bicocca. His research focuses on models and algorithms for data analysis and decision making under uncertainty in the areas of Business Intelligence, Data and Text Mining and Computational Finance. In the winter term 2015/16 he is giving the course 625.605 – Business Intelligence in Klagenfurt for the second time.

Prof. Tadeusz Morzy | 23.10.2015 | 14:00 Uhr | E.2.69

Abstract

Many applications require processing and analyzing sequential data. Examples include the analysis of passenger traveling histories, stock market prices, purchases of customers over time, meteorological events, workflow and RFID logs, etc. Recently, issues related to warehousing and analytical processing (OLAP) of sequential data have received growing attention. Particularly, the concept of Sequence OLAP (SOLAP) has been proposed that support OLAP processing of different kinds of aggregate queries on sequential data. The main feature distinguishing SOLAP from traditional OLAP is that data sequences managed by an SOLAP system are characterized by subsequence patterns they possess. The SOLAP systems allow to group data sequences based on patterns they possess and apply aggregate functions to each group. This kind of SOLAP queries are called a sequential pattern-based aggregate (PBA) queries. The processing of PBA queries is expensive due to the fact that they require full scan of all stored sequences. The natural question is how to efficiently evaluate this kind of queries?

The talk focuses on the new index structure supporting processing of sequential pattern-based aggregate queries. The structure of the index will be presented as well as classes of pattern-based aggregate queries supported by the index will be discussed. Finally, the performance of PBA queries using the proposed index will be presented.

Tadeusz Morzy is a professor in the Computing Science Department of Poznan University of Technology. He received his M. Sc., Ph. D. and Polish Habilitation from the Technical University of Poznań, Poland. He has held visiting positions at the Loyola University, New Orleans, Klagenfurt University Austria, University La Sapienza Italy, and Free University Amsterdam. He has authored and coauthored over 100 papers on databases, data mining, and data warehousing. He is co-author of a book on „Concurrency Control in Distributed Database Systems“ by North-Holland, editor and coauthor of “Handbook on Data Management” by Springer, and author of “Data Mining: Methods and Algorithms” (in Polish). He served as General Chair of the 2nd and 16th ADBIS Conferences (1998, 2012), and has served/serves on numerous program committees of international conferences and workshops. His research interests include data mining, data warehousing, transaction processing in database and data warehouse systems, access methods and query processing for databases, database optimization and performance evaluation.