Rule based expert system

An expert system is software that attempts to provide an answer to a

problem, or clarify uncertainties where normally one or more
human experts would need to be consulted. Expert systems are most
common in a specific problem domain, and is a traditional application
and/or subfield of artificial intelligence (AI). A wide variety of methods can
be used to simulate the performance of the expert; however, common to
most or all are:

1. The creation of a knowledge base which uses some knowledge

representation structure to capture the knowledge of the Subject

Matter Expert (SME);

2. a process of gathering that knowledge from the SME and codifying it

according to the structure, which is called knowledge engineering; and

3. Once the system is developed, it is placed in the same real

world problem solving situation as the human SME, typically as an aid to
human workers or as a supplement to some information system. Expert
systems may or may not have learning components

rule-based expert systems, use human expert knowledge to solve real-

world problems that normally would require human intelligence. Expert
knowledge is often represented in the form of rules or as data within the
computer.

Depending upon the problem requirement, these rules and data can be
recalled to solve problems. Rule-based expert systems have played an
important role in modern intelligent systems and their applications in
strategic goal setting, planning, design, scheduling, fault monitoring,
diagnosis and so on.
Complete structure of a rule-based
expert system
External
Database External Program

Expert System
Knowledge Base Database

Rule: IF-THEN Fact

Inference Engine

Explanation Facilities

Developer
User Interface Interface

User
Knowledge Engineer

Expert
Characteristics of rule based expert
system
 An expert system is built to perform at a human expert level
in a narrow, specialised domain. Thus, the most important
characteristic of an expert system is its high-quality
performance. No matter how fast the system can solve a
problem, the user will not be satisfied if the result is wrong.
 On the other hand, the speed of reaching a solution is very
important. Even the most accurate decision or diagnosis
may not be useful if it is too late to apply, for instance, in an
emergency, when a patient dies or a nuclear power plant
explodes.
 Expert systems apply heuristics to guide the reasoning and
thus reduce the search area for a solution.
 A unique feature of an expert system is its explanation
capability. It enables the expert system to review its own
reasoning and explain its decisions.
 Expert systems employ symbolic reasoning when solving a
problem. Symbols are used to represent different types of
knowledge such as facts, concepts and rules

Advantage of rule based expert system

 Separation of knowledge from its processing:

The structure of a rule-based expert system provides an effective

separation of the knowledge base from the inference engine. This

makes it possible to develop different applications using the same

expert system shell.

 Dealing with incomplete and uncertain knowledge

 Most rule-based expert systems are capable of representing and
reasoning with incomplete and uncertain knowledge

Disadvantage of rule based expert system

 Opaque relations between rules:

Although the individual production rules are relatively simple and

self-documented, their logical interactions within the large set of
rules may be opaque. Rule-based systems make it difficult to
observe how individual rules serve the overall strategy.

 Ineffective search strategy:

The inference engine applies an exhaustive search through all the
production rules during each cycle. Expert systems with a large set of
rules (over 100 rules) can be slow, and thus large rule-based systems
can be unsuitable for real-time applications

The McCulloch-Pitts Model of Neuron

The early model of an artificial neuron is introduced by Warren McCulloch and
Walter Pitts in 1943. The McCulloch-Pitts neural model is also known as linear
threshold gate. It is a neuron of a set of inputs and one I 1, I 2, I 3 , … .. I m output Y . The
linear threshold gate simply classifies the set of inputs into two different classes.
Thus the output Y  is binary. Such a function can be described mathematically using
these equations:

(2.1)
 (2.2)

W 1 , W 2 ,W 3 … … W m are weight values normalized in the range of either (0,1) or (−1,1) 

and associated with each input line,   is the weighted sum, and   is a threshold
constant. The function  f  is a linear step function at threshold   as shown in
figure 2.3. The symbolic representation of the linear threshold gate is shown in
figure 2.4 [Has95].

Figure 2.3: Linear Threshold Function

Figure 2.4: Symbolic Illustration of Linear Threshold Gate

The McCulloch-Pitts model of a neuron is simple yet has substantial computing

potential. It also has a precise mathematical definition. However, this model is so
simplistic that it only generates a binary output and also the weight and threshold
values are fixed. The neural computing algorithm has diverse features for various
applications [Zur92]. Thus, we need to obtain the neural model with more flexible
computational features

Cover’s theorem on separability of patterns

“A complex pattern-classification problem cast in high-dimensional space

nonlinearly is more likely to be linearly separable than in a low dimensional
space”

(Cover, 1965).

 Let X denote a set of N patterns (points) x1,x2,x3,…,xN

 Each point is assigned to one of two classes: X+ and X-

 This dichotomy is separable if there exist a surface that separates these two
classes of points.

 For each pattern define the next

vector: φ ( x )=[φ1 ( x ) , φ2 ( x ) , … … … … .. φM ( x ) ]T

 The vector φ (x) maps points in a p-dimensional input space into

corresponding points in a new space of dimension m.

 Each φ i( x ) is a hidden function, i.e., a hidden unit

 A dichotomy {X+,X-} is said to be φ-separable if there exist a m-dimensional

vector w such that we may write (Cover, 1965):

 wT φ(x) ≥ 0, x X+

 wT φ(x) < 0, x X-

 The hyperplane defined by wT φ(x) = 0, is the separating surface between the

two classes.
  Given a set of patterns X in an input space of arbitrary dimension p, we can
usually find a nonlinear mapping φ(x) of high enough dimension M such
that we have linear separability in the φ space.

2.4.1 Linear Separability and the XOR Problem

Consider two-input patterns ( X 1 , X 2 ) being classified into two classes as shown in
figure 2.9. Each point with either symbol of   or   represents a pattern with a set of
values ( X 1 , X 2 ). Each pattern is classified into one of two classes. Notice that these
classes can be separated with a single line  . They are known as linearly
separable patterns. Linear separability refers to the fact that classes of patterns
with  -dimensional vector  X =(x 1 , x 2 , … … ., x n) can be separated with a
single decision surface. In the case above, the line   represents the decision surface.

Figure 2.9:linearly separable patterns

The processing unit of a single-layer perceptron network is able to categorize a set of

patterns into two classes as the linear threshold function defines their linear
separability. Conversely, the two classes must be linearly separable in order for the
perceptron network to function correctly . Indeed, this is the main limitation of a
single-layer perceptron network.

The most classic example of linearly inseparable pattern is a logical exclusive-OR

(XOR) function. Shown in figure 2.10 is the illustration of XOR function that two
classes, 0 for black dot and 1 for white dot, cannot be separated with a single line. The
solution seems that patterns of ( x 1 , x 2) can be logically classified with two lines  L1 
and  L2 

Figure 2.10:Exclusive-0R fuction

Comparision of MLP and RBFN

Multilayer perceptrons (MLPs) and radial basis function (RBF) networks are

the two most commonly-used types of feedforward network. They have much
more in common than most of the NN literature would suggest. The only
fundamental difference is the way in which hidden units combine values coming
from preceding layers in the network--MLPs use inner products, while RBFs use
Euclidean distance. There are also differences in the customary methods for
training MLPs and RBF networks, although most methods for training MLPs can
also be applied to RBF networks. Furthermore, there are crucial differences
between two broad types of RBF network--ordinary RBF networks and normalized

RBF networks--that are ignored in most of the NN literature. These differences

have important consequences for the generalization ability of the networks,
especially when the number of inputs is large.

 Multilayer perception stands for a neural network with one or

more hidden layer.

 Properties of multilayer neural networks:

 o The model of each neuron in the network includes a
nonlinear activation function that’s differentiable.

o Network contains one or more hidden layer.

o Network exhibits a high degree of connectivity through its

synaptic weights.

 Common deficiencies in multilayer neural networks:

o Theoretical analysis of MLNN is difficult to undertake.

 This comes from the nonlinearity and high connectivity

of the network.

o Harder efforts are required for visualizing the learning

process.

 This comes from the existence of several layers in the

network

Advantage and disadvantage of RBFN

• RBF trains faster than a MLP

• Another advantage that is claimed is that the hidden layer is easier

to interpret than the hidden layer in an MLP.

Although the RBF is quick to train, when training is finished and it is

being used it is slower than a MLP, so where speed is a factor a MLP
may be more appropriate.

T-Norm Operator:-
T-norm (also T-norm or, unabbreviated, triangular norm) is a kind of binary
operation used in the framework of probabilistic metric spaces and in multi-
valued logic, specifically in fuzzy logic. A t-norm generalizes intersection in
a lattice and conjunction in logic. The name triangular norm refers to the
fact that in the framework of probabilistic metric spaces t-norms are used to
generalize triangle inequality of ordinary metric spaces.

Definition :-

 T-norm is a function T: [0, 1] × [0, 1] → [0, 1] which satisfies the following

properties:

 Commutativity: T(a, b) = T(b, a)

 Monotonicity: T(a, b) ≤ T(c, d) if a ≤ c and b ≤ d
 Associativity: T(a, T(b, c)) = T(T(a, b), c)
 The number 1 acts as identity element: T(a, 1) = a

Since a t-norm is a binary algebraic operation on the interval [0, 1], infix

algebraic notation is also common, with the t-norm usually denoted by * .

The defining conditions of the t-norm are exactly those of the partially
ordered Abelian monoid on the real unit interval [0, 1]. (Cf. ordered
group.) The monoidal operation of any partially ordered Abelian monoid L is
therefore by some authors called a triangular norm on L.
 Minimum t-norm   also called the Gōdel t-norm, as it is
the standard semantics for conjunction in Gōdel fuzzy logic. Besides that, it occurs in most t-norm
based fuzzy logics as the standard semantics for weak conjunction. It is the pointwise largest t-norm
(see the properties of t-norms below).
Graph of the product t-norm

 Product t-norm  (the ordinary product of real numbers). Besides

other uses, the product t-norm is the standard semantics for strong conjunction in product fuzzy logic.
It is a strict Archimedean t-norm.

 Łukasiewicz t-norm   The name comes

from the fact that the t-norm is the standard semantics for strong conjunction in Łukasiewicz fuzzy
logic. It is a nilpotent Archimedean t-norm, pointwise smaller than the product t-norm.

 Drastic t-norm

The name reflects the fact that the drastic t-norm is the pointwise smallest t-norm (see
the properties of t-norms below). It is a right-continuous Archimedean t-norm.

 Nilpotent minimum

is a standard example of a t-norm which is left-continuous, but not continuous. Despite its name,
the nilpotent minimum is not a nilpotent t-norm.

 Hamacher product

is a strict Archimedean t-norm, and an important representative of the parametric classes

of Hamacher t-norms and Schweizer–Sklar t-norms.