After completing this course successfully, the student can:
Describe and compare machine learning methods and techniques, associated complexity and the application domain. Theories and techniques include: Bayesian decision theory, expectation maximization algorithms,decision trees, linear discrimination, neural networks, hidden Markov models, boosting and performance evaluation.
Explain the theoretical aspect of machine learning techniques such as VC dimension, and bias/variance dilemma.
Design, implement and systematically evaluate machine learning methods and models.
Apply machine learning methods and techniques in a realistic case (taken from research projects) in the area of Human Media Interaction such as Human Behavior, Pose Recognition and Brain Computer Interfacing.
The first part of the course is an introduction to the theory and practicalities of machine learning, in which the emphasis will be on an overview of the various techniques, their associated complexity and the application domain. We will also look into the theoretical aspects of machine learning techniques, such as VC dimension, over- and underfitting and the Bias/Variance Dilemma.
Assignments will be based on real-life cases to give students the opportunity to apply the knowledge and techniques they have acquired to practical problems. Emphasis will be on methodology (how to achieve reliable models systematically) and the evaluation of the learnt/trained models.
The second, more practical part of the course will be geared towards the retrieval of knowledge and the ability to apply advanced models, especially on online learning and the classification and prediction of time series, as these techniques are particularly important in Human Media Interaction (HMI). Students will work in pairs on a realistic case (taken from actual research projects) and will have to work methodologically through the whole procedure from cleaning up data to the final model selection and performance evaluation.
The emphasis of these projects will be on machine learning in HMI such as Human Behavior, Pose Recognition and machine learning techniques in Brain Computer Interfacing. A concrete example of is training a Kinect to recognize poses. This course has no final exam: the grade will be based on assignments, progress presentations and the report of the project.
Working knowledge in probability theory and statistics. This is covered in the course Probability Theory and Statistics (191530082).
Knowledge about basic AI formalisms for defining and solving problems: search, representation of knowledge, reasoning, learning and reasoning under uncertainty. This is covered in the bachelors course Artificial Intelligence (192140302).
Machine Learning, Supervised Learning, Bayesian Decision Theory, Parametric Models, Decision Trees, Linear Discrimination, Neural Networks, Committee of Classifiers, Evaluation of Classifiers.
|Voorkennis noodzakelijk: Kansrekening & Statistiek (191530082), Artificial Intelligence (192140302)||Verplicht materiaal|
|E. Alpaydin: Introduction to Machine Learning (second edition), MIT-Press 2010, ISBN 978-0-262-01243-0|
|Assignments and final project|