Fundamentals of Artificial Intelligence VI

The course “Fundamentals of Artificial Intelligence VI” introduces the mathematical foundations of supervised learning and deals in detail with two learning algorithms that are widely used in practice: decision tree learning and neural networks.

Participants will learn how to apply these algorithms to practical problems and how to assess the quality of a trained classifier. The course concludes with a discussion of the risks and challenges of machine learning.

The course is based on the textbook by Stuart Russell and Peter Norvig: Introduction to Artificial Intelligence, 3rd edition, 2012 or 4th edition, 2020. The 3rd edition is also available in German: Stuart Russell and Peter Norvig: Künstliche Intelligenz – Ein Moderner Ansatz, 3rd revised edition, Pearson 2012.

The course is held in English with German subtitles.

Module Introduction to Machine Learning 

• The learning agent – a review
• Types of learning
• Occam's razor 

Module Decision Tree Learning 

• Boolean decision trees
• Restaurant example
• Overfitting
• Theoretical properties
• Recursive decision tree learning algorithm
• Information gain and entropy
• Entropy of a boolean variable
• Information gain
• Use in decision tree learning 

Module Neural Networks 

• Introduction
• History of neural networks
• McCulloch-Pitts "unit" model of a neuron
• Activation functions
• Feedforward networks
• Single-layer feedforward networks
• Single-layer feedforward networks
• Theoretical properties and limitations
• Backpropagation algorithm
• Example: turning right 

Module Evaluation, Challenges and Risks 

• Process of machine learning
• Training and test data
• Learning curve
• K-fold cross-validation
• Evaluation of a boolean classifier
• False positive and false negative results
• Accuracy, sensitivity, specificity, precision
• F1 score
• Risks and challenges of machine learning
• Statistical vs. causal models
• Incorrect predictions with high confidence
• Universal adversarial perturbations 

At the end of the course, you will be able to

• understand the theoretical foundations of machine learning algorithms,
• evaluate a trained classifier,
• apply decision tree learning to practical problems,
• apply neural network learning to practical problems,
• understand the challenges and risks of machine learning algorithms

None.