Page 169 - DCAP506_ARTIFICIAL_INTELLIGENCE

Unit 12: Learning

Decision Trees Notes

 This is a third strategy to concept learning.
 To categorize a specific input, we begin at the top of the tree and reply questions until we
reach a leaf, where the specification is stored.

 ID3 is a program example for Decision Trees.
 ID3 utilizes iterative method to construct decision trees, favoring simple trees over
complex ones, on the theory that simple trees are more precise classifiers of future
inputs.
 It starts by selecting a random subset of the training examples.
 This subset is known as the window.

 The algorithm constructs a decision tree that accurately categorizes all examples in the
window.

Task Illustrate the steps used in Winston’s Program.

Self Assessment

Fill in the blanks:
6. The thought of generating a classification program that can develop its own class definitions
is known as ......................... .
7. The objective of ......................... is to generate a description that is reliable with all positive
examples but no negative examples in the training set.

12.5 Explanation-based Learning

An Explanation-based Learning (EBL) system accepts an example (i.e. a training example)
and illustrates what it learns from the example. The EBL system takes only the pertinent
features of the training. This clarification is converted into specific form that a problem solving
program can understand. The explanation is generalized so that it can be utilized to solve other
problems.
PRODIGY is a system that incorporates problem solving, planning, and learning methods in a
single design. It was formerly envisioned by Jaime Carbonell and Steven Minton, as an AI
system to test and build up ideas on the role that machine learning plays in planning and
problem solving. PRODIGY utilizes the EBL to obtain control rules.

The EBL module utilizes the results from the problem-solving trace (i.e. Steps in solving
problems) that were produced by the central problem solver (a search engine that searches over
a problem space). It builds explanations by means of an axiomatized theory that illustrates both
the domain and the architecture of the problem solver. The results are then converted as control
rules and added to the knowledge base. The control knowledge that comprises control rules is
utilized to direct the search process efficiently.

LOVELY PROFESSIONAL UNIVERSITY 163