perClass Course: Modules content
Module 1: Introduction to Pattern Recognition.
Pattern recognition approach; concept of a class; labeled/unlabeled observations; training from examples; data representation; generalization; optimization. Representaion by features; constructing data sets; handling data and meta-data information; visualization using scatter and image views; normalization issues; data visualization and its relevance.
Module 2: Supervised Learning.
Bayes theorem; generative and discriminative classifiers; parametric and non-parametric models; naive Bayes; linear, quadratic, and mixture models; Parzen density estimation; linear discriminant analysis; logistic classifier; nearest-neighbor rule; support-vector machines; perceptron; neural networks; decision trees; decision operating points; construction of detectors.
Module 3: Evaluation
Error and performance measures; confusion matrix; learning curves; overtraining; classifier complexity; cross-validation; ROC analysis for two-class and multi-class problems; rejection of outliers and in the area of class overlap; class imbalance; handling of prior probabilities.
Module 4: Dimensionality reduction
Per-feature class distributions; measures of overlap; feature selection; individual, greedy, and floating search; filter and wrapper criteria; genetic search; feature extraction; PCA, LDA, non-linear extraction methods. Representing measurements by proximities; building classifier in dissimilarity spaces; training similarity measures from examples (for image/signal data).
Module 5: Advanced Pattern Recognition
System design work-flow. How to define a problem? How to design and optimize system? cascades and hierarchical classifiers; selecting operating points; cost-optimzation; cross-validation over objects; classification performance vs execution speed trade-off; implementing and evaluating custom algorithms (dimensionality reduction + model + ROC optimization of operating point).
Module 6: Unsupevised Learning: Cluster Analysis.
How to define groups of similar observations. Interpreting the clusters; using clusters to build better classifiers in multi-modal problems; visualizing clustering solutions; leveraging clustering as a tool to understand the source of classification errors; deciding on the number of clusters;
Tools: Dissimilarity measures; k-means; mixture models, EM algorithm; hierarchical clustering; evidence accumulation
Module 7: Classifier embedding
Embedding classifiers in production; perClass Runtime; exporting trained classifiers outside Matlab; linking perClass Runtime to a custom application; accessing decision names; using multiple pipelines; changing operating points in production; strategies to speed up classifier execution; performance vs speed characteristics