Inhalt

•  
  Introduction
•  
   Motivation
•  
  Data and representation
•  
   Workflow pipeline for machine learning applications
•  
   Challenges of complex data
•  
   Feature-based representation
•  
   Dissimilarity-based representation
•  
   Other types of data representation
•  
  Thesis overview
•  
   Scientific contributions
•  
   Structural overview
•  
   Publications and funding in the context of this thesis
•  
  Tools for supervised and unsupervised learning with dissimilarity data
•  
   Motivation
•  
   Scientific contributions and structure of the chapter
•  
  Relational learning vector quantization
•  
   Introduction
•  
   Generalized learning vector quantization
•  
   Pseudo-Euclidean embedding of dissimilarity data
•  
   GLVQ for dissimilarity data
•  
   Reducing computational demand via Nyström approximation
•  
   Interpretability of relational prototypes
•  
   Experiments
•  
   Concluding remarks
•  
  Relational generative topographic mapping
•  
   Introduction
•  
   Generative topographic mapping (GTM)
•  
   Relational GTM
•  
   Experiments
•  
   Concluding remarks
•  
  Adaptive metrics for complex data
•  
   Motivation
•  
   Scientific contributions and structure of the chapter
•  
   Vector-based metric learning in LVQ
•  
   Motion tracking data
•  
   Proof-of-concept example
•  
   Sequence alignment as a parameterized dissimilarity measure
•  
   Learning scoring parameters from labeled data
•  
  Practical implementation
•  
   Algorithm overview
•  
   Meta-parameters
•  
   Proof-of-concept with artificial data
•  
   RGLVQ error function surface
•  
   Influence of crispness on the alignment
•  
  Experiments with real-world data
•  
   Experimental procedure
•  
   Copenhagen Chromosomes
•  
   Intelligent tutoring systems for Java programming
•  
   Reducing computational demand
•  
   Discussion
•  
  Unsupervised suitability assessment for data representations
•  
   Motivation
•  
   Scientific contributions and structure of the chapter
•  
   Low-dimensional Euclidean embeddings
•  
  Quantitative quality assessment
•  
   Principles of quality assessment for DR
•  
   Evaluating DR based on the co-ranking matrix
•  
   Point-wise quality measure
•  
   Parameterization of the quality measure
•  
   Experiments with real-world data
•  
   Discussion
•  
  Comparing dissimilarity-based representations
•  
   Introduction
•  
   How to compare dissimilarity measures?
•  
   Comparison of metrics for the Euclidean vector space
•  
   Comparison of non-Euclidean settings
•  
   Discussion
•  
   Conclusion
•  
  Additional information
•  
   Derivative of soft alignment
•  
   Information about the Chromosomes data set
•  
   Information about the Sorting data set
•  
   Publications in the context of this thesis
•  
   Bibliography