Learning by Analogy, Inductive Learning, Explanation Based Learning

Learning by Analogy:

It is a powerful mechanism that is used for exploiting the past experience in planning and problem solving. This process is developed to extract knowledge from past successful problem-solving situations that bear a strong similarity to the current problem. The analogy is one of the central inference methods in human cognition as well as a powerful computational mechanism. An analogical transformation process is developed to extract knowledge from past problem.

While encountering a new problem situation a person is reminded of past situations that holds strong similarity to the present problem. The given type of reminding experience serves to retrieve behaviors that were appropriate in earlier problem-solving sessions whereupon past behavior is adapted to meet the demands of the current situation. The commonality among the previous and the current situations as well as the successful applications of modified plans can serve as the basis for generalization.

While the humans exhibit a universal ability to learn from experience no matter what the task is AI systems are seldom designed to model this adaptive quality. There are two fundamental hypotheses and they are:

• The hypothesis I: Problem-solving and learning are the inalienable aspects of a unified cognitive mechanism. In other words, one cannot acquire the requisite cognitive skills without solving problems and the very process of solving problems provides the information which is necessary to acquire and tune the problem-solving skills. The second hypothesis postulated a unified learning mechanism.
• Hypothesis II: The same learning mechanism that accounts for concept formation in declarative domains operate in acquiring problem-solving skills and formulating the generalized plans.

One method of verifying the second hypothesis is to develop a problem-solving mechanism into which one can integrate the techniques involved in concept.

Inductive Learning:

It is known to be the simplest form to learn a function from examples f where f is the target function. An example is a pair (x, f(x)).

Problem: Find a hypothesis h such that h ≈ f for a given training set of examples. This is a highly simplified model of real learning.

• Ignores prior knowledge.
• Assumes a deterministic and observable environment.
• Assumes examples are given.
• Assume that the agent wants to learn m f (why?).

Constructor adjusts h to agree with f on training set where h is consistent if it agrees with f on all examples. For example curve fitting:

Constructor adjusts h to agree with f on training set where h is consistent if it agrees with f on all examples. For example curve fitting:

Constructor adjusts h to agree with f on training set where h is consistent if it agrees with f on all examples. For example curve fitting:

Explanation-based learning:

An Explanation-based Learning (EBL) system accepts a training example and explains what it learns from the example. EBL system takes only the relevant aspects of training. This explanation is translated into the particular form that a problem-solving program can understand. The explanation is generalized in such a way that it can be used to solve other problems as well.

The PRODIGY is a system that integrates problem-solving, planning and learning methods in a single architecture. It was originally conceived by Jaime Carbonell and Steven Minton as an AI system to test and develop the ideas on the role that machine learning plays in planning and problem solving. PRODIGY uses the EBL to acquire the control rules.

The EBL module uses the results from the problem-solving trace (that is steps in solving problems) that were generated by the central problem solver (a search engine that searches over a problem space). It constructs the explanations using an axiomatized theory that describes both the domain and the architecture of the problem solver. The results are then translated into the control rules and added to the knowledge base. The control knowledge that contains control rules, is used to guide the search process effectively.

References:

1. Elaine Rich, Kevin Knight 1991, "Artificial Intelligence".
2. Nilsson, Nils J. Principles of Artificial Intelligence, Narosa Publishing House New Delhi, 1998.
3. Norvig, Peter & Russel, Stuart Artificial Intelligence: A modern Approach, Prentice Hall, NJ, 1995
4. Patterson, Dan W. Introduction to Artificial Intelligence and Expert Systems, Prentice Hall of India Private Limited New Delhi, 1998.