Main content

Top content

Foundations of Machine Learning: Theory and Practice

8.30807

Dozenten

Prof. Dr. Elia Bruni

Beschreibung

This seminar provides an introduction to the foundations of machine learning, covering essential concepts, algorithms, and techniques in traditional machine learning. If there will be time left, we wll quickly look at basic Neural Networks, but that is not guaranteed.

Participants will gain a deep understanding of supervised, unsupervised, and semi-supervised learning, as well as insights into statistical machine learning and key optimization methods. The course emphasizes both theory and practical implementation, preparing students for advanced machine learning applications and research.

Machine Learning Algorithms' Categories:
Supervised, unsupervised, and semi-supervised learning, classification, regression, clustering, parametric vs. non-parametric, and linear vs. nonlinear approaches.

Supervised Learning:
Linear regression, logistic regression, support vector machines, linear discriminant analysis, decision trees, ensemble methods (random forest, boosting, XGBoost).

Optimization and Loss Functions:
Gradient descent, variations (SGD, Momentum, RMSprop, ADAM), logistic loss, cross-entropy, hinge loss.

Model Evaluation and Selection:
Metrics (accuracy, precision, recall, F-score, ROC curve, AUC), confusion matrix, handling imbalanced datasets, cross-validation.

Unsupervised Learning:
Clustering (k-means, DBSCAN, hierarchical clustering), Gaussian Mixture Models, dimension reduction techniques (PCA, ICA, t-SNE).

Bayesian algorithms:
Naive Bayes, Maximum a posteriori (MAP) estimation, Maximum Likelihood (ML) estimation

Statistical significance:
R-squared, P-values

Data Handling:
Missing data, imbalanced data, data distribution shifts

Similarity/dissimilarity metrics:
Euclidean, Manhattan, Cosine, Mahalanobis (advanced)

Weitere Angaben

Ort: 01/B01
Zeiten: Di. 14:00 - 16:00 (wöchentlich)
Erster Termin: Dienstag, 15.04.2025 14:00 - 16:00, Ort: 01/B01
Veranstaltungsart: Seminar (Offizielle Lehrveranstaltungen)