Statistical Pattern Recognition
Prof. Thomas BroxStatistical pattern recognition, nowadays often known under the term "machine learning", is the key element of modern computer science. Its goal is to find, learn, and recognize patterns in complex data, for example in images, speech, biological pathways, the internet. In all these cases we would like to relate abstract symbols, such as the word "dog", to patterns that cannot be described by a simple equation or a set of logical statements. Can you describe the symbol "dog" by just using the pixel values of an image? How do you deal with the variation among dog images? Can you recognize an unknown dog as a dog? What is the pattern that is common to all dogs? Statistical pattern recognition tries to answer these questions by looking at many examples, the more the better. This is very different from classical computer science where the computer program, the algorithm, is the key element of the process. In machine learning we have a learning algorithm, but the actual information is not in the algorithm but in the data processed and stored by this algorithm.
This course gives an introduction in the classical tasks of pattern recognition: classification, regression, and clustering. In the case of classification, we learn a decision function from annotated training examples (e.g., a set of dog images). Given a new image, the classifier should be able to tell whether it is a dog image or not. In regression we learn a mapping from an input function to an output function. Again this mapping is learned from a set of input/output pairs. Both classification and regression are supervised methods as the data comes together with the correct output. Clustering is an unsupervised learning method, where we are just given unlabeled data and where clustering should separate the data into reasonable subsets. The course is based in large parts on the textbook "Pattern Recognition and Machine Learning" written by Christopher Bishop. The exercises will consist of theoretical assignments and programming assignments in Matlab.
The content of this course is complementary to the Machine Learning course offered by Joschka Boedecker and Frank Hutter. It absolutely makes sense to attend both courses if you want to specialize in Machine Learning.
|
|
Slides and Recordings
Additional information and some helpful hints regarding the exercises can be found here.
Recordings from Olaf Ronneberger (2015):
| Class 1: | Introduction |
| Class 2: | Probability distributions |
| Class 3: | Mixture models, clustering, and EM |
| Class 4: | Nonparametric methods |
| Class 5: | Regression |
| Class 7: | Classification |
| Class 8: | Support vector machines |
| Class 9: | Deep Learning |
| Class 10: | Projection methods |
| Class 12: | Sampling methods |