Syllabus

1. Introduction.
Definition of learning systems. Goals and applications of machine learning
(classication and regression). Basics on statistical learning theory (Vapnik
Chervonenkis bound). Undertting and Overfitting. Use of data: training set,
test set, validation set. (4 classes)

2. Articial Neural Networks.
Neurons and biological motivation.  The Perceptron
and its learning algorithm. (2 classes)
Multi-Layer Feedforward Neural Networks.  Back-
propagation (BP) algorithm. BP batch version and BP on-line version. Momentum updating rule. (4 classes)
Radial-Basis function (RBF) networks: regularized and generalized RBF net-
works. Learning strategies and
error functions. Unsupervised selection of center. Supervised selection of
weights and centers: decomposition methods into two blocks and decomposi-
tion methods into more blocks.  (4 classes)

3. Support Vector Machines (Kernel methods)
Soft and hard Maximum Margin Classiers. Quadratic programming formula-
tion of the soft/hard maximum margin separators. Kernels methods. KKT conditions (2 classes)
Dual formulation of the primal QP problem. Wolfe duality theory for QP.
 Decomposition methods. (4 classes)
Implementation tricks: Caching, shrinking. (2 classes)
Multiclass SVM problems: one-against-one and one-against-all. (2 classes). Choosing parameters: k-fold cross-validation. (2 classes)

4. Practical use of learning algorithms. Comparing learning algorithms from the optimization point of view.  Use of standard software (4 classes)

 

Download the pdf version of the syllabus