Section 3 : Logic and knowledge in learning

Commentary

Section Goals

• To combine knowledge representation with learning methods.
• To introduce learning methods that can take advantage of prior knowledge about the world.

Learning Objectives

Learning Objective 1

• Outline the hypothesis space represented in logic formation.
• Describe the hypothesis space search methods and algorithms, such as the current-best-hypothesis search.
• Explain the version space learning method, and its related algorithm.
• Describe the general schemes of knowledge-related learning methods, such as EBL, RBL, and KBIL in the form of hypothesis and entailment constraint.
• Explain the following concepts or terms:
  • Prior knowledge
  • Generalization
  • Specialization
  • Version space
  • Candidate elimination
  • Dropping conditions
  • Generalization hierarchy
  • Entailment constraint
  • Explanation-based learning (EBL)
  • Relevance-based learning (RBL)
  • Knowledge-based inductive learning (KBIL)

Objective Readings

Required readings:

Reading topics:

Logical Formation of Learning, Hypothesis and Version Space Search, General Schemes of Knowledge in Learning (see Sections 19.1-19.2 of AIMA3ed)

Objective Questions

• What are the benefits of using prior knowledge in learning?
• How are generalization and specialization of hypotheses used in hypothesis search and construction?

Objective Activities

• Explore the following algorithms related to this section in the textbook (AIMA3ed):
  • Current-Best-Learning
  • Version-Space-Learning
• Complete Exercise 19.2 of AIMA3ed.