A5074196124- Data Management and Algorithms
- Advanced Clustering Algorithms Research
- Advanced Database Systems and Queries
- Time Series Analysis and Forecasting
- Data Mining Algorithms and Applications
- Anomaly Detection Techniques and Applications
- Advanced Image and Video Retrieval Techniques
- Image Retrieval and Classification Techniques
- Algorithms and Data Compression
- Gene expression and cancer classification
- Isotope Analysis in Ecology
- Image Processing and 3D Reconstruction
- Face and Expression Recognition
- Geographic Information Systems Studies
- Automated Road and Building Extraction
- Archaeology and ancient environmental studies
- Data Visualization and Analytics
- Complex Network Analysis Techniques
- Semantic Web and Ontologies
- Advanced Graph Neural Networks
- Pleistocene-Era Hominins and Archaeology
- Biomedical Text Mining and Ontologies
- Bioinformatics and Genomic Networks
- Forensic Anthropology and Bioarchaeology Studies
- Bayesian Methods and Mixture Models
Kiel University
2021-2025
Clinical Research Center Kiel
2024
Ludwig-Maximilians-Universität München
2012-2022
LMU Klinikum
2006-2020
Institut für Urheber- und Medienrecht
2003-2014
RWTH Aachen University
2010
As a prolific research area in data mining, subspace clustering and related problems induced vast quantity of proposed solutions. However, many publications compare new proposition—if at all—with one or two competitors, even with so-called “naïve” ad hoc solution, but fail to clarify the exact problem definition. consequence, if solutions are thoroughly compared experimentally, it will often remain unclear whether both tackle same or, they do, agree certain tacit assumptions how such may...
Abstract Clustering refers to the task of identifying groups or clusters in a data set. In density‐based clustering , cluster is set objects spread space over contiguous region high density objects. Density‐based are separated from each other by regions low Data located low‐density typically considered noise outliers. © 2011 John Wiley & Sons, Inc. WIREs Mining Knowl Discov 1 231–240 DOI: 10.1002/widm.30 This article categorized under: Technologies > Structure Discovery and
Many outlier detection methods do not merely provide the decision for a single data object being or an but give also score "outlier factor" signaling "how much" respective is outlier. A major problem any user very acquainted with method in question how to interpret this "factor" order decide numeric again whether indeed Here, we formulate local density based providing "score" range of [0, 1] that directly interpretable as probability
Several application domains such as molecular biology and geography produce a tremendous amount of data which can no longer be managed without the help efficient effective mining methods. One primary tasks is clustering. However, traditional clustering algorithms often fail to detect meaningful clusters because most real-world sets are characterized by high dimensional, inherently sparse space. Nevertheless, contain interesting hidden in various subspaces original feature Therefore, concept...
Outlier scores provided by different outlier models differ widely in their meaning, range, and contrast between and, hence, are not easily comparable or interpretable. We propose a unification of various translation the arbitrary “outlier factors” to values range [0, 1] interpretable as describing probability data object being an outlier. As application, we show that this facilitates enhanced ensembles for detection.
In this paper, we propose a novel outlier detection model to find outliers that deviate from the generating mechanisms of normal instances by considering combinations different subsets attributes, as they occur when there are local correlations in data set. Our enables search for arbitrarily oriented subspaces original feature space. We show how addition an score, our also derives explanation outlierness is useful investigating results. experiments suggest method can than existing work and...
Abstract Clustering refers to the task of identifying groups or clusters in a data set. In density‐based clustering , cluster is set objects spread space over contiguous region high density objects. Density‐based are separated from each other by regions low Data located low‐density typically considered noise outliers. this review article we discuss statistical notion clusters, classic algorithms for deriving flat partitioning methods hierarchical clustering, and semi‐supervised clustering....
The detection of correlations between different features in a set feature vectors is very important data mining task because correlation indicates dependency the or some association cause and effect them. This can be arbitrarily complex, i.e. one more might dependent from combination several other features. Well-known methods like principal components analysis (PCA) perfectly find which are global, linear, not hidden noise vectors, uniform, same type exhibited all vectors. In many...
Many clustering algorithms tend to break down in high-dimensional feature spaces, because the clusters often exist only specific subspaces (attribute subsets) of original space. Therefore, task projected (or subspace clustering) has been defined recently. As a solution tackle this problem, we propose concept local preferences, which captures main directions high point density. Using concept, adopt density-based cope with data. In particular, achieve following advantages over existing...
Subspace clustering has been investigated extensively since traditional algorithms often fail to detect meaningful clusters in high-dimensional data spaces. Many recently proposed subspace methods suffer from two severe problems: First, the typically scale exponentially with dimensionality and/or of clusters. Second, for performance reasons, many use a global density threshold clustering, which is quite questionable subspaces significantly different will most likely exhibit varying...
The reverse k-nearest neighbor (RkNN) problem, i.e. finding all objects in a data set the neighbors of which include specified query object, is generalization 1-nearest problem has received increasing attention recently. Many industrial and scientific applications call for solutions RkNN arbitrary metric spaces where are not Euclidean only distance function given specifying object similarity. Usually, these need solution generalized value k known advance may change from to query. However,...
Previous chapter Next Full AccessProceedings Proceedings of the 2012 SIAM International Conference on Data Mining (SDM)Density-based Projected Clustering over High Dimensional StreamsIrene Ntoutsi, Arthur Zimek, Themis Palpanas, Peer Kröger, and Hans-Peter KriegelIrene Kriegelpp.987 - 998Chapter DOI:https://doi.org/10.1137/1.9781611972825.85PDFBibTexSections ToolsAdd to favoritesExport CitationTrack CitationsEmail SectionsAboutAbstract high dimensional data streams is an important problem in...
Mobility data captures the locations of moving objects such as humans, animals, and cars. With availability Global Positioning System (GPS)–equipped mobile devices other inexpensive location-tracking technologies, mobility is collected ubiquitously. In recent years, use has demonstrated a significant impact in various domains, including traffic management, urban planning, health sciences. this article, we present domain science. Towards unified approach to science, pipeline having following...
Abstract Subspace clustering refers to the task of identifying clusters similar objects or data records (vectors) where similarity is defined with respect a subset attributes (i.e., subspace space). The not necessarily (and actually usually not) same for different within one solution. In this article, problems motivating are sketched, definitions and usages subspaces described, exemplary algorithmic solutions discussed. Finally, we sketch current research directions. © 2012 Wiley...
Abstract The production of metal forming technology products until now requires substantial expertise from specialists in product, process, and equipment design. Particularly important is the ability to compensate for stochastic, unpredictable process deviations. Given this context, a newly established Priority Program 2422 (PP 2422) by German Research Foundation (DFG) aims enhance current FEM-simulation-based design active surfaces tools with data-driven modeling. present paper firstly...
Abstract. In the framework of a changing climate, it is useful to devise methods capable effectively assessing and monitoring landscape air–sea CO2 fluxes. this study, we developed an integrated machine learning tool objectively classify track marine carbon biomes under seasonally interannually environmental conditions. The was applied monthly output global ocean biogeochemistry model at 0.25° resolution run atmospheric forcing for period 1958–2018. Carbon are defined as regions having...