IT 407 - Knowledge Engineering 1

LECTURE NOTES
ON
IT407 KNOWLEDGE ENGINEERING

MODULE 4

Prepared By
Prof. SHIJU KUMAR P S
Department of Information Technology
M E S COLLEGE OF ENGINEERING, KUTTIPPURAM

MESCE, DEPT OF IT 1
 IT 407 - Knowledge Engineering 2

KNOWLEDGE ENGINEERING SYLLABUS

Learning: What is learning, Rote learning, Learning by Taking Advice, Learning in Problem-
solving, Learning from example: induction, Explanation-based learning.

Module 4
1. What is Learning?
“If an expert system--brilliantly designed, engineered and implemented--cannot learn not to
repeat its mistakes, it is not as intelligent as a worm or a sea anemone or a kitten.”

-- Oliver G. Selfridge, from The Gardens of Learning

"Find a bug in a program, and fix it, and the program will work today. Show the program how
to find and fix a bug, and the program will work forever."

-- Oliver G. Selfridge, from AI’s Greatest Trends and Controversies

Machine learning refers to a system capable of the autonomous acquisition and integration of
knowledge. This capacity to learn from experience, analytical observation, and other means,
results in a system that can continuously self-improve and thereby offer increased efficiency
and effectiveness.

Types of Learning:
• Rote Learning
• Learning by Taking Advice
• Learning in Problem Solving (Reinforcement Learning)
• Learning from Examples: Induction (Instance-Based Learning)
• Explanation-Based Learning
• Discovery
• Analogy
• Neural Net and Genetic Learning

2. Rote Learning

It avoids understanding the inner complexity but focuses on memorizing the materials so that
it can be recalled exactly the way it was read.

• Learning by Memorization- It avoid inner complexities and focuses on memorizing

material so that it can be recalled later for the same inputs.
• Learning by something by repeating- In this “Over and over again” method is used. it
is similar to know we learn poems.

Rote learning is basically Memorization

1. Saving knowledge so it can be used again

2. Retrieval is the only problem
3. No repeated computation, inference or query necessary. Eg: Caching

MESCE, DEPT OF IT 2
 IT 407 - Knowledge Engineering 3

When a computer stores a piece of data, it is performing a rudimentary form of learning. In

the case of data caching, we store computed values—such as strongly successful paths in
search—so that we do not have to recompute them later. When an agent receives an experience
episode from another agent and remembers it, it learns.

Research issues in this type of learning include organization of information for efficient
updates and retrieval, and generalization of information for effective updates and retrieval, as
well as interpretation and analysis.

3. Learning by Taking Advice

• Computer can do very little without program.

• When programs write series of programs a rudimentary kind of learning takes
place.
• After being programed the computer can do something which could not do it
before.
• Interpreter or compiler is needed to intervene. To change the teacher’s
instruction in to code that machine can execute directly.
• Direct instruction on how to respond to situation.

Request → Interpret → Operationalize → Integrate → Evaluate

This type of learning is natural in humans. However, a system that performs this type of
learning must be able to translate advice given into actionable decisions. Thus, in this case,
the system might not be doing any learning at all, as humans probably are the ones who do that
for the system. Research issues in this area involve efficient representation of advice and
effective parsing of advice to “operationalize” them.

For example, we know that “Defense wins championships” is a piece of good advice. But we
need to translate this high-level strategy into actionable tactics, such as playing “man-to-man
defense”, “substitute often”, etc.

4. Learning in Problem Solving

• No teacher advices but learning by experience.

• It does not involve increase in knowledge just method of using knowledge.

There are three basic methods in which a system can learn from its own experiences.
• Learning by Parameter Adjustment
• Learning by Macro Operators
• Learning by Chunking

4.1 Learning by parameter Adjustment.

Many programs rely on an evaluation procedure to summaries the state of search etc. Game
playing programs provide many examples of this.

MESCE, DEPT OF IT 3
 IT 407 - Knowledge Engineering 4

However, many programs have a static evaluation function.

In learning a slight modification of the formulation of the evaluation of the problem is required.

Here the problem has an evaluation function that is represented as a polynomial of the form
such as:

The t terms a value of features and the c terms are weights.

In designing programs, it is often difficult to decide on the exact value to give each weight
initially.

So, the basic idea of idea of parameter adjustment is to:

• Start with some estimate of the correct weight settings.

• Modify the weight in the program on the basis of accumulated experiences.
• Features that appear to be good predictors will have their weights increased and bad
ones will be decreased.

Samuel's Checkers programs employed 16 such features at any one time chosen from a pool of
38.

4.2 Learning by Macro Operators

• The basic idea here is similar to Rote Learning:

Avoid expensive recomputation

• Macro-operators can be used to group a whole series of actions into one.

For example: Making dinner can be described a lay the table, cook dinner, serve dinner. We
could treat laying the table as on action even though it involves a sequence of actions.

• The STRIPS problem-solving employed macro-operators in it's learning phase.

• Consider a blocks world example in which ON(C,B) and ON(A,TABLE) are true.

STRIPS can achieve ON(A,B) in four steps:

• UNSTACK(C,B), PUTDOWN(C), PICKUP(A), STACK(A,B)

• STRIPS now builds a macro-operator MACROP with preconditions ON(C,B),
ON(A,TABLE), postconditions ON(A,B), ON(C,TABLE) and the four steps as its
body.
• MACROP can now be used in future operation.
• But it is not very general. The above can be easily generalized with variables used in
place of the blocks.

However, generalization is not always that easy (See Rich and Knight).

MESCE, DEPT OF IT 4
 IT 407 - Knowledge Engineering 5

4.3 Learning by Chunking

• Chunking involves similar ideas to Macro Operators and originates from psychological
ideas on memory and problem solving.
• The computational basis is in production systems (studied earlier).
• SOAR is a system that use production rules to represent its knowledge. It also employs
chunking to learn from experience.

Basic Outline of SOAR's Method

• SOAR solves problems it fires productions these are stored in long term memory.
• Some firings turn out to be more useful than others.
• When SOAR detects are useful sequence of firings, it creates chunks.
• A chunk is essentially a large production that does the work of an entire sequence of
smaller ones.
• Chunks may be generalised before storing.

5. Learning from Examples: Induction

Often times, we learn by classifying things into different labels and then “inferring” the
common characteristics of each label. This type of learning is especially useful in classification
tasks: the “learned” or “trained” system can classify a previously unseen entity (or instance)
into one of the group by comparing the instance’s attribute values to the common
characteristics of the groups.

From another perspective, this type of learning can be viewed as learning the decision path to
correctly label an instance. And that decision path can be represented as part of a decision tree,
where a node is a decision point, branches are possible values of a decision, and leaves are
classification labels.

Whereas reinforcement learning dominates intelligent, real-time, adaptive systems, induction

(aka instance-based learning) dominates classification systems. Usually, a developer is given
a set of training instances such that each instance has a feature vector and a class label. The
developer then runs his or her system—with an implemented learning algorithm—on the set of
instances until the system is able to classify all instances in the set meeting a required level of
accuracy. Then the system is ready for use.

Research issues in this type of learning are wide ranging: feature selection, active learning,
handling noise and incomplete data, size of the training set, size of the testing set, classification
efficiency, and so forth.

6. Explanation-Based Learning

Explanation-based learning (EBL) is based on the notion of single-example learning. Recall

that instance-based learning requires typically a substantial number of training instances.
However, (1) it is possible that we do not have or afford such a number of training instances,
and (2) it is at times possible to learn well from a single example, especially when we have
enough knowledge.

MESCE, DEPT OF IT 5
 IT 407 - Knowledge Engineering 6

Unlike instance-based learning which is more data-intensive, data-driven, EBL is more

knowledge-intensive, knowledge-driven.

First, an EBL system accepts a training example. Given its goal concept, an operationality
criterion (a description of which concepts are usable), and domain theory (which are usually a
set of rules that describe relationships between objects and actions in a domain), it
“generalizes” the training example to describe the goal concept and to satisfy the operationality
criterion.

Research issues in this area have involved applications to theorem provers, and knowledge
acquisition and engineering.

MESCE, DEPT OF IT 6