@article{BlanchardFlaskaHandyetal.2016, author = {Blanchard, Gilles and Flaska, Marek and Handy, Gregory and Pozzi, Sara and Scott, Clayton}, title = {Classification with asymmetric label noise: Consistency and maximal denoising}, series = {Electronic journal of statistics}, volume = {10}, journal = {Electronic journal of statistics}, publisher = {Institute of Mathematical Statistics}, address = {Cleveland}, issn = {1935-7524}, doi = {10.1214/16-EJS1193}, pages = {2780 -- 2824}, year = {2016}, abstract = {In many real-world classification problems, the labels of training examples are randomly corrupted. Most previous theoretical work on classification with label noise assumes that the two classes are separable, that the label noise is independent of the true class label, or that the noise proportions for each class are known. In this work, we give conditions that are necessary and sufficient for the true class-conditional distributions to be identifiable. These conditions are weaker than those analyzed previously, and allow for the classes to be nonseparable and the noise levels to be asymmetric and unknown. The conditions essentially state that a majority of the observed labels are correct and that the true class-conditional distributions are "mutually irreducible," a concept we introduce that limits the similarity of the two distributions. For any label noise problem, there is a unique pair of true class-conditional distributions satisfying the proposed conditions, and we argue that this pair corresponds in a certain sense to maximal denoising of the observed distributions. Our results are facilitated by a connection to "mixture proportion estimation," which is the problem of estimating the maximal proportion of one distribution that is present in another. We establish a novel rate of convergence result for mixture proportion estimation, and apply this to obtain consistency of a discrimination rule based on surrogate loss minimization. Experimental results on benchmark data and a nuclear particle classification problem demonstrate the efficacy of our approach.}, language = {en} } @article{RichterDoellner2014, author = {Richter, Rico and D{\"o}llner, J{\"u}rgen Roland Friedrich}, title = {Concepts and techniques for integration, analysis and visualization of massive 3D point clouds}, series = {Computers, environment and urban systems}, volume = {45}, journal = {Computers, environment and urban systems}, publisher = {Elsevier}, address = {Oxford}, issn = {0198-9715}, doi = {10.1016/j.compenvurbsys.2013.07.004}, pages = {114 -- 124}, year = {2014}, abstract = {Remote sensing methods, such as LiDAR and image-based photogrammetry, are established approaches for capturing the physical world. Professional and low-cost scanning devices are capable of generating dense 3D point clouds. Typically, these 3D point clouds are preprocessed by GIS and are then used as input data in a variety of applications such as urban planning, environmental monitoring, disaster management, and simulation. The availability of area-wide 3D point clouds will drastically increase in the future due to the availability of novel capturing methods (e.g., driver assistance systems) and low-cost scanning devices. Applications, systems, and workflows will therefore face large collections of redundant, up-to-date 3D point clouds and have to cope with massive amounts of data. Hence, approaches are required that will efficiently integrate, update, manage, analyze, and visualize 3D point clouds. In this paper, we define requirements for a system infrastructure that enables the integration of 3D point clouds from heterogeneous capturing devices and different timestamps. Change detection and update strategies for 3D point clouds are presented that reduce storage requirements and offer new insights for analysis purposes. We also present an approach that attributes 3D point clouds with semantic information (e.g., object class category information), which enables more effective data processing, analysis, and visualization. Out-of-core real-time rendering techniques then allow for an interactive exploration of the entire 3D point cloud and the corresponding analysis results. Web-based visualization services are utilized to make 3D point clouds available to a large community. The proposed concepts and techniques are designed to establish 3D point clouds as base datasets, as well as rendering primitives for analysis and visualization tasks, which allow operations to be performed directly on the point data. Finally, we evaluate the presented system, report on its applications, and discuss further research challenges.}, language = {en} } @article{WestphalAxelssonNeuhausetal.2014, author = {Westphal, Florian and Axelsson, Stefan and Neuhaus, Christian and Polze, Andreas}, title = {VMI-PL: A monitoring language for virtual platforms using virtual machine introspection}, series = {Digital Investigation : the international journal of digital forensics \& incident response}, volume = {11}, journal = {Digital Investigation : the international journal of digital forensics \& incident response}, publisher = {Elsevier}, address = {Oxford}, issn = {1742-2876}, doi = {10.1016/j.diin.2014.05.016}, pages = {S85 -- S94}, year = {2014}, abstract = {With the growth of virtualization and cloud computing, more and more forensic investigations rely on being able to perform live forensics on a virtual machine using virtual machine introspection (VMI). Inspecting a virtual machine through its hypervisor enables investigation without risking contamination of the evidence, crashing the computer, etc. To further access to these techniques for the investigator/researcher we have developed a new VMI monitoring language. This language is based on a review of the most commonly used VMI-techniques to date, and it enables the user to monitor the virtual machine's memory, events and data streams. A prototype implementation of our monitoring system was implemented in KVM, though implementation on any hypervisor that uses the common x86 virtualization hardware assistance support should be straightforward. Our prototype outperforms the proprietary VMWare VProbes in many cases, with a maximum performance loss of 18\% for a realistic test case, which we consider acceptable. Our implementation is freely available under a liberal software distribution license. (C) 2014 Digital Forensics Research Workshop. Published by Elsevier Ltd. All rights reserved.}, language = {en} } @phdthesis{Huebner2007, author = {H{\"u}bner, Sebastian Valentin}, title = {Wissensbasierte Modellierung von Audio-Signal-Klassifikatoren : zur Bioakustik von Tursiops truncatus. - 2., {\"u}berarb. Aufl.}, url = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus-16631}, school = {Universit{\"a}t Potsdam}, year = {2007}, abstract = {Die vorliegende Arbeit befasst sich mit der wissensbasierten Modellierung von Audio-Signal-Klassifikatoren (ASK) f{\"u}r die Bioakustik. Sie behandelt ein interdisziplin{\"a}res Problem, das viele Facetten umfasst. Zu diesen geh{\"o}ren artspezifische bioakustische Fragen, mathematisch-algorithmische Details und Probleme der Repr{\"a}sentation von Expertenwissen. Es wird eine universelle praktisch anwendbare Methode zur wissensbasierten Modellierung bioakustischer ASK dargestellt und evaluiert. Das Problem der Modellierung von ASK wird dabei durchg{\"a}ngig aus KDD-Perspektive (Knowledge Discovery in Databases) betrachtet. Der grundlegende Ansatz besteht darin, mit Hilfe von modifizierten KDD-Methoden und Data-Mining-Verfahren die Modellierung von ASK wesentlich zu erleichtern. Das etablierte KDD-Paradigma wird mit Hilfe eines detaillierten formalen Modells auf den Bereich der Modellierung von ASK {\"u}bertragen. Neunzehn elementare KDD-Verfahren bilden die Grundlage eines umfassenden Systems zur wissensbasierten Modellierung von ASK. Methode und Algorithmen werden evaluiert, indem eine sehr umfangreiche Sammlung akustischer Signale des Großen T{\"u}mmlers mit ihrer Hilfe untersucht wird. Die Sammlung wurde speziell f{\"u}r diese Arbeit in Eilat (Israel) angefertigt. Insgesamt werden auf Grundlage dieses Audiomaterials vier empirische Einzelstudien durchgef{\"u}hrt: - Auf der Basis von oszillographischen und spektrographischen Darstellungen wird ein ph{\"a}nomenologisches Klassifikationssystem f{\"u}r die vielf{\"a}ltigen Laute des Großen T{\"u}mmlers dargestellt. - Mit Hilfe eines Korpus halbsynthetischer Audiodaten werden verschiedene grundlegende Verfahren zur Modellierung und Anwendung von ASK in Hinblick auf ihre Genauigkeit und Robustheit untersucht. - Mit einem speziell entwickelten Clustering-Verfahren werden mehrere Tausend nat{\"u}rliche Pfifflaute des Großen T{\"u}mmlers untersucht. Die Ergebnisse werden visualisiert und diskutiert. - Durch maschinelles mustererkennungsbasiertes akustisches Monitoring wird die Emissionsdynamik verschiedener Lauttypen im Verlaufe von vier Wochen untersucht. Etwa 2.5 Millionen Klicklaute werden im Anschluss auf ihre spektralen Charakteristika hin untersucht. Die beschriebene Methode und die dargestellten Algorithmen sind in vielf{\"a}ltiger Hinsicht erweiterbar, ohne dass an ihrer grundlegenden Architektur etwas ge{\"a}ndert werden muss. Sie lassen sich leicht in dem gesamten Gebiet der Bioakustik einsetzen. Hiermit besitzen sie auch f{\"u}r angrenzende Disziplinen ein hohes Potential, denn exaktes Wissen {\"u}ber die akustischen Kommunikations- und Sonarsysteme der Tiere wird in der theoretischen Biologie, in den Kognitionswissenschaften, aber auch im praktischen Naturschutz, in Zukunft eine wichtige Rolle spielen.}, language = {de} } @phdthesis{Dornhege2006, author = {Dornhege, Guido}, title = {Increasing information transfer rates for brain-computer interfacing}, url = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus-7690}, school = {Universit{\"a}t Potsdam}, year = {2006}, abstract = {The goal of a Brain-Computer Interface (BCI) consists of the development of a unidirectional interface between a human and a computer to allow control of a device only via brain signals. While the BCI systems of almost all other groups require the user to be trained over several weeks or even months, the group of Prof. Dr. Klaus-Robert M{\"u}ller in Berlin and Potsdam, which I belong to, was one of the first research groups in this field which used machine learning techniques on a large scale. The adaptivity of the processing system to the individual brain patterns of the subject confers huge advantages for the user. Thus BCI research is considered a hot topic in machine learning and computer science. It requires interdisciplinary cooperation between disparate fields such as neuroscience, since only by combining machine learning and signal processing techniques based on neurophysiological knowledge will the largest progress be made. In this work I particularly deal with my part of this project, which lies mainly in the area of computer science. I have considered the following three main points: Establishing a performance measure based on information theory: I have critically illuminated the assumptions of Shannon's information transfer rate for application in a BCI context. By establishing suitable coding strategies I was able to show that this theoretical measure approximates quite well to what is practically achieveable. Transfer and development of suitable signal processing and machine learning techniques: One substantial component of my work was to develop several machine learning and signal processing algorithms to improve the efficiency of a BCI. Based on the neurophysiological knowledge that several independent EEG features can be observed for some mental states, I have developed a method for combining different and maybe independent features which improved performance. In some cases the performance of the combination algorithm outperforms the best single performance by more than 50 \%. Furthermore, I have theoretically and practically addressed via the development of suitable algorithms the question of the optimal number of classes which should be used for a BCI. It transpired that with BCI performances reported so far, three or four different mental states are optimal. For another extension I have combined ideas from signal processing with those of machine learning since a high gain can be achieved if the temporal filtering, i.e., the choice of frequency bands, is automatically adapted to each subject individually. Implementation of the Berlin brain computer interface and realization of suitable experiments: Finally a further substantial component of my work was to realize an online BCI system which includes the developed methods, but is also flexible enough to allow the simple realization of new algorithms and ideas. So far, bitrates of up to 40 bits per minute have been achieved with this system by absolutely untrained users which, compared to results of other groups, is highly successful.}, subject = {Kybernetik}, language = {en} }