Chapter 11: Analytical Learning

• Inductive learners perform poorly when insufficient data is available
• Prior knowledge can reduce the complexity of the hypothesis space
• In inductive learning, the learner learns a hypothesis that is consistent with the training examples
• In analytical learning, the learner learns a hypothesis that is consistent with both the training examples and the background knowledge
• In this chapter, we will assume that the background knowledge (the domain theory) is correct

An Example

Take a look at Table 11.1, that illustrates SafeToStack(x,y)

Explanation Based Learning - Table 11.2

1. Explain the training example - Figure 11.2
2. Analyze the explanation - Figure 11.3

Comments

• EBL performs knowledge compilation by reformulating the domain theory
• The proof process can result in the discovery of new features (similar to the hidden nodes in a neural network)
• The inductive bias of EBL is the domain theory plus a preference for small sets of maximally general Horn clauses - thus, the bias is not directly a function of the EBL method itself
• Knowledge level learning can still occur. Knowledge level learning occurs when the learned hypotheses are not entailed by the domain theory alone. For example, learning that Moroccans speak Arabic from the domain theory that all people in a country speak a common language and the specific example of a Moroccan speaking Arabic.

Exercises

• 11.1(a)