In this course, we study the fundaments of pattern recognition. We adopt an engineering point of view on the development of intelligent machines which are able to identify patterns in data. The core methods and algorithms are elaborated that enable pattern recognition for a wide range of data sources including sensory data (image, video, audio, location, etc.) as well as born-digital data (text, network traffic, chemical formulas, etc.). The course is organized in two parts. In the first part, we explore statistical pattern recognition based on feature vector representation. Standard methods for unsupervised clustering and supervised classification in vector spaces will be discussed. In the second part, we investigate structural pattern recognition based on string and graph representation. For clustering and classification of structural data, dissimilarity measures will be introduced alongside with explicit and implicit vector space embedding approaches. The course is accompanied by practical exercises that involve the implementation of algorithms discussed in class and their application to exemplary pattern recognition tasks.

On successful completion of this class, you will be able to:

- Design pattern recognition systems for a large variety of data sources, namely to cluster and classify objects represented as feature vectors, feature vector sequences, strings, and graphs.

- Describe the mathematical techniques, assumptions, and relevant parameters of the underlying recognition algorithms, including k-means clustering, Bayes classification, support vector machines, neural networks, hidden Markov models, graph edit distance, and graph kernel functions.

- Apply the pattern recognition systems to exemplary recognition tasks ranging from image recognition over movement analysis to the classification of molecular compounds.