What does a Learning Agent look like?
Taxi example for a learning agent?
What is an atomic representation of the states of the world?
states e.g. identified with labels (names)
like a “black box”, “image”;
-> search algorithms may treat states in this way
How does factored representation represent states of the “world”?
attributes (variables) with values exploited
e.g. in planning, constraint satisfaction
How does structured respresentation represent states of the “world”?
dependence model between attributes
e.g. logical knowledge bases, Bayesian networks
What is unsupervised learning?
no explicit feedback
• most common: clustering of “similar” examples, concept formation
What is reinforcement learning?
occasional rewards or punishments in terms of payoff
What is supervised learning? What is semi supervised learning?
What is overfitting? How is it solved?
What is Decision Tree Pruning?
Prevent split on “irrelevant” attributes, using information gain and statistical methods
(Avoid overfitting)
What is cross validation?
Use part of the data for training, part for testing
-> avoid overfitting
What are the three typical manifestations of learning curves? Under which circumstances do they appear?
realizable: the function can be learned
unrealizable: function can’t be learned
(missing attributes/ hypothesis space too restrictive)
redundant: too many irrelevant attributes
Last changed2 years ago
