Lehrmethodik inkl. Einsatz von eLearning-Tools
Lectures, practical exercises, and an optional project possibly chosen by the student and a topic of the student's choice.
Inhalt/e
Reinforcement learningReinforcement learning is about making sequences of decisionsStunning achievements of reinforcement learningHow to find good sequences of decisions in an unknown domain through exploration and learning?Delayed rewards, long-term benefits of decisions, exploration and exploitationImproving decision policy through explorationGeneralizing what has been learnedLearning from examples and background knowledgeHow to use prior knowledge in Machine Learning?Learning in logic – Inductive Logic Programming (ILP)Algorithms for learning programs from examples in ILPDiscovering new abstract conceptsLearning qualitative models with applications in roboticsHow to model qualitatively, avoiding numbers?Reasoning and simulation with qualitative modelsLearning qualitative models from observationsLearning and planning of robot tasks: rescue robot, cart-pole balancing, humanoid robot, quadcopterLearning from noisy dataProblems with noise in learning dataKey ideas to cope with noise: simpler models are often betterAlgorithms for learning decision trees from noisy dataHow to estimate probabilities in machine learning correctly?Argument-Based Machine Learning (ABML)Human expert may help learning by annotating training examples with argumentsAn algorithm for learning rules from argumented examplesDiscovering problem structure with function decompositionThe idea of structuring the learning problem with function decompositionDiscovering structure with HINT algorithmImproving accuracy and interpretability by structure learning