http://www.inf.tu-dresden.de/index.php?node_id=438&ln=en Talk announcements of the Faculty of Computer Science at the Technische Universität Dresden en-us Mon, 20 May 2013 12:05:52 GMT 60 http://blogs.law.harvard.edu/tech/rss webmaster@inf.tu-dresden.de http://www.inf.tu-dresden.de/index.php?node_id=438&ln=de In this news feed, you find information about presentations at the department. Thu, 16 Feb 2006 17:10:56 GMT www.inf.tu-dresden.de/en/talk/welcome http://www.inf.tu-dresden.de/index.php?node_id=438&ln=en#t1138 13th Jun 2013, 1.00 PM, INF 1004 (Ratssaal), "Machine Learning based Test Process Control": PhD defence by Dipl.-Inf. Matthias KirmseA growing complexity in modern semiconductor production requires more and more elaborate test processes. As a result, test process faults and corresponding test errors become increasingly frequent and expensive. Thereby, studies have shown that conventional models often fail to efficiently detect, diagnose and recover these faults. In our thesis, we present new machine learning based approaches for each of these main test process control areas. We provide extensive experimental results underlying their ability to significantly decrease test process fault related cost. Moreover, we present a productive test error detection system based on our new approach that has been successfully applied in the studied test department for over two years. Originating from these semiconductor studies, we furthermore developed the novel meta learning approach “Large Margin Rectangle Learning” (LMRL). It combines the interpretability of hyperrectangle based models and the minimal risk property of the large margin principle. Besides introducing its theoretical background, we provide empirical evidence for the supposed margin-accuracy relation. And we finally present experimental results showing that LMRL outperforms a majority of the compared machine learning approaches, especially the studied interpretable methods. Altogether, LMRL is a promising new approach to create more accurate interpretable models. Fri, 17 May 2013 07:02:27 GMT www.inf.tu-dresden.de/en/talk/1138 http://www.inf.tu-dresden.de/index.php?node_id=438&ln=en#t1137 30th May 2013, 1.00 PM, INF 1004 (Ratssaal), "Frequent Itemset Mining on Multiprocessor Systems": PhD defence by Dipl.-Inf. Benjamin SchlegelFrequent-itemset mining is an important building block in many data mining applications like market basket analysis, recommendation, web-mining, fraud detection, and gene expression analysis. In many of them, the datasets being mined can easily grow up to hundreds of giga- bytes or even terabytes of data. Hence, efficient algorithms are required to process such large amounts of data. In recent years, there have been many frequent-itemset mining algorithms proposed, which however (1) often have high memory requirements and (2) do not exploit the large degrees of parallelism provided by modern multiprocessor systems. The high mem- ory requirements arise mainly from inefficient data structures that have only been shown to be sufficient for small datasets. For large datasets, however, the use of these data structures force the algorithms to go out-of-core, i.e., they have to access secondary memory, which leads to serious performance degradations. Exploiting available parallelism is further required to mine large datasets because the serial performance of processors almost stopped increasing. Algorithms should therefore exploit (1) the large number of available threads and (2) also the other kinds of parallelism (e.g., vector instruction sets) besides thread-level parallelism. In this work, we tackle the high memory requirements of frequent-itemset mining twofold: we (1) compress the datasets being mined because they must be kept in main memory during several mining invocations and (2) improve existing mining algorithms with memory-efficient data structures. For compressing the datasets, we employ efficient encodings that show a good compression performance on a wide variety of realistic datasets, i.e., the size of the datasets is reduced by up to 6.4x. The encodings can further be applied directly while loading the dataset from disk or network. Since encoding and decoding is repeatedly required for loading and mining the datasets, we reduce its costs by providing parallel encodings that achieve high throughputs for both tasks. For a memory-efficient representation of the mining algorithms’ intermediate data, we propose compact data structures and even employ explicit compression. Both methods together reduce the intermediate data’s size by up to 25x. The smaller memory requirements avoid or delay expensive out-of-core computation when large datasets are mined. For coping with the high parallelism provided by current multiprocessor systems, we iden- tify the performance hot spots and scalability issues of existing frequent-itemset mining al- gorithms. The hot spots, which form basic building blocks of these algorithms, cover (1) counting the frequency of fixed-length strings, (2) building prefix trees, (3) compressing in- teger values, and (4) intersecting lists of sorted integer values or bitmaps. For all of them, we discuss how to exploit available parallelism and provide scalable solutions. Furthermore, almost all components of the mining algorithms must be parallelized to keep the sequential fraction of the algorithms as small as possible. We integrate the parallelized building blocks and components into three well-known mining algorithms and further analyze the impact of certain existing optimizations. Our algorithms are already single-threaded often up an order of magnitude faster than existing highly optimized algorithms and further scale almost linear on a large 32-core multiprocessor system. Although our optimizations are intended for fre- quent itemset mining algorithms, they can be applied with only minor changes to algorithms that are used for mining of other types of itemsets. Thu, 16 May 2013 13:16:56 GMT www.inf.tu-dresden.de/en/talk/1137 http://www.inf.tu-dresden.de/index.php?node_id=438&ln=en#t1135 24th May 2013, 1.00 PM, INF 3105 (Beratungsraum, 3. Etage), "Identifizierbarkeit von RFID-Tags": Presentation of the student project (IST) by Sebastian HerzbergImmer mehr Gegenstände werden mit Funktechnologien ausgestattet, die sich ganz allgemein dem Bereich „RFID“ zuordnen lassen. Bei diesen Gegenständen handelt es sich unter anderem um gewöhnliche RFID-Tags aber auch Smart Cards, Ausweise, Smart Phones, Fahrkarten etc. Da viele dieser Gegenstände einen starken Personenbezug haben und von Menschen mit sich herumgetragen werden, stellt sich die Frage, inwiefern eine Wiedererkennbarkeit von Gegenständen, die mit RFID-Funktechnologie ausgestattet sind, möglich ist, um auf diese Weise letztlich den betreffenden Menschen wiederzuerkennen. Diese Arbeit soll einen Überblick über bekannte Ansätze zur Wiedererkennbarkeit bezüglich der RFID-Funktechnologie geben. Dabei soll nicht nur der durch digital vorliegende Daten mögliche Informationsgewinn (Sicherungsschicht und höher), sondern insbesondere auch der durch analoge Signale mögliche Informationsgewinn (Bitübertragungsschicht) betrachtet werden. Neben dem Zusammentragen der aus der Literatur bekannten Erkenntnisse sollen einige der vorgefundenen Aussagen exemplarisch durch eigene Untersuchungen nachvollzogen werden. Wünschenswert wäre, wenn aus der Literatur bekannte Identifizierungsmöglichkeiten im Rahmen dieser Untersuchungen verbessert würden. Tue, 14 May 2013 08:40:06 GMT www.inf.tu-dresden.de/en/talk/1135