Meeting 22: Introduction to Supervised Learning

Reading: AIAMA 18.1-3

One of the most challenging parts of agent design is building the model, since it needs to be sophisticated enough to capture the salient aspects of the environment. A large class of methods to assist with model construction is to use learning. Learning can be used either as a black box approach, i.e. learn the percept to action mapping directly or through learning of parts of the agent model, often through some form of parameter settings. In this meeting we will introduce the notion of learning and see how they fit into agent design and adaptation. We will then look at supervised learning.

Questions you should be able to answer after reading are:

  1. What compenents of an agent architecture is learning applied?
  2. What is the difference between inductive and deductive learning?
  3. What is the difference between supervised and unsupervised learning?
  4. How does reinforcement learning differ from supervised learning?
  5. What is the difference between a training set and a test set in supervised learning? When is each used?
  6. What does a hypothesis refere to in learning theory?
  7. What does it mean for a system to generalize well?
  8. What is the difference between regression and classification problems?
  9. Why should we believe Ockham's razor is a good principle to follow?
  10. What is a decision tree?