CS 485 - Statistical and Computational Foundations of Machine Learning

Extracting meaningful patterns from random samples of large data sets. Statistical analysis of the resulting problems. Common algorithmic paradigms for such tasks. Central concepts: VC-dimension, margins of a classifier, sparsity and description length, other types of regularization. Performance guarantees: generalization bounds, data dependent error bounds, and computational complexity of learning algorithms. Common paradigms: neural networks, kernel methods and support-vector machines, boosting, nearest neighbor classifiers. Applications to data mining.