UHH: Prof. Dr. Ulrike von Luxburg, Dr. Tobias Lang
- Sheet 01
Ex 1: Dataset Exploration - vaccination of children: cond. probabilities,
refuting claim of vaccination causing allergy (similar to Simpson's paradox.
Ex 2: Writing and testing a kNN-classifier with 0-1-loss with different k-values. Evaluation of performance.
- Sheet 02
Ex 1: kNN-classification on handwritten digits (USPS dataset) - classfication and performance evaluation.
Ex 2: kNN-text-classification of user posts in 20 newsgroups. Optimization of kNN implementation via binomic
formula and loop avoidance via matrix operations.
Ex 3: Bayes classification
Ex 4: Letter classification
Ex 5: Types of errors: Example with mail classification
- Sheet 03
Ex 1: Linear mapping
Ex 2: Empirically investigate the accuracy of such binomic models
Ex 3: Least-squares regression: Implementing, learning, evaluation, non-linear features and outliers.
Ex 4: Decision boundary for a 2D mix of Gaussians.
- Sheet 04
Ex 1: Ridge regression: implementation, testing and L2-loss.
Ex 2: Prediction complexity of linear least square and kNN: Comparison
Ex 3: Convexity: Proof that L2 loss is convex.
Ex 4: Inverse of a matrix: Proof
Ex 5: Least-squares regression for multi-class classification: Iris dataset
Ex 6: Decision boundary: Example with 1NN and linear least-squares regression
without and with quadratic basis functions.
- Sheet 05
Ex 1: Understanding of SVMs: Different kernels, different parameters.
Ex 2: Understanding of SVMs 2: RBF and polynomial kernel, differen parameters.
Ex 3: Kernel SVM: Example
Ex 4: Building new kernels: Proof of properties.
Ex 5: Polynomial kernel: feature map, dimension
Ex 6: SVM cancer detection: (soft margin) SVM with cross validation.
- Sheet 06
Ex 1: PCA: direction of principal components - example
Ex 2: Variance: Survey dataset with question regarding PCA
Ex 3: PCA: implementation and testing.
Ex 4: PCA on USPS data (handwritten digits)
Ex 5: MapReduce with Matlab
- Sheets 07, 08, 09
Sheet 07: Perform an unsupervised data analysis study.
Sheet 08: Perform a predictive modeling study.
Sheet 09: Grade reports of fellow students.