UNIT 3 FUZZY SYSTEMS

Introduction to fuzzy- Fuzzy sets and Fuzzy reasoning - Fuzzy matrices - Fuzzy
Functions -Decomposition - Fuzzy rules and inferences - Fuzzy decision making -
deep learning architecture for Fuzzy logic.

Fuzzy Logic
The 'Fuzzy' word means the things that are not clear or are vague. Sometimes,
we cannot decide in real life that the given problem or statement is either true or false.
At that time, this concept provides many values between the true and false and gives
the flexibility to find the best solution to that problem.
Fuzzy logic contains the multiple logical values and these values are the truth
values of a variable or problem between 0 and 1.
This concept was introduced by Lofti Zadeh in 1965 based on the Fuzzy Set
Theory. This concept provides the possibilities which are not given by computers, but
similar to the range of possibilities generated by humans. In the Boolean system, only
two possibilities (0 and 1) exist, where 1 denotes the absolute truth value and 0 denotes
the absolute false value.
But in the fuzzy system, there are multiple possibilities present between the 0
and 1, which are partially false and partially true. The Fuzzy logic can be implemented
in systems such as micro-controllers, workstation-based or large network-based
systems for achieving the definite output. It can also be implemented in both hardware
or software.
Characteristics of Fuzzy Logic
1. This concept is flexible and we can easily understand and implement it.
2. It is used for helping the minimization of the logics created by the human.
3. It is the best method for finding the solution of those problems which are suitable for
approximate or uncertain reasoning.
4. It always offers two values, which denote the two possible solutions for a problem
and statement.
5. It allows users to build or create the functions which are non-linear of arbitrary
complexity.
6. In fuzzy logic, everything is a matter of degree.
7. In the Fuzzy logic, any system which is logical can be easily fuzzified.
8. It is based on natural language processing.
9. It is also used by the quantitative analysts for improving their algorithm's execution.
10. It also allows users to integrate with the programming.

Architecture of a Fuzzy Logic System

In the architecture of the Fuzzy Logic system, each component plays an important role.
The architecture consists of the different four components which are given below.
Rule Base
Fuzzification
Inference Engine
Defuzzification
1. Rule Base

Rule Base is a component used for storing the set of rules and the If-Then conditions
given by the experts are used for controlling the decision-making systems. There are so
many updates that come in the Fuzzy theory recently, which offers effective methods
for designing and tuning of fuzzy controllers. These updates or developments are
decreases the number of fuzzy set of rules.
2. Fuzzification
Fuzzification is a module or component for transforming the system inputs, i.e., it
converts the crisp number into fuzzy steps. The crisp numbers are those inputs which
are measured by the sensors and then fuzzification passed them into the control systems
for further processing. This component divides the input signals into following five
states in any Fuzzy Logic system:
Large Positive (LP)
Medium Positive (MP)
Small (S)
Medium Negative (MN)
Large negative (LN)

3. Inference Engine
This component is a main component in any Fuzzy Logic system (FLS), because all the
information is processed in the Inference Engine. It allows users to find the matching
degree between the current fuzzy input and the rules. After the matching degree, this
system determines which rule is to be added according to the given input field. When
all rules are fired, then they are combined for developing the control actions.
4. Defuzzification
Defuzzification is a module or component, which takes the fuzzy set inputs generated
by the Inference Engine, and then transforms them into a crisp value. It is the last step
in the process of a fuzzy logic system. The crisp value is a type of value which is
acceptable by the user. Various techniques are present to do this, but the user has to
select the best one for reducing the errors.
Advantages of Fuzzy Logic
1. The methodology of this concept works similarly as the human reasoning.
2. Any user can easily understand the structure of Fuzzy Logic.
3. It does not need a large memory, because the algorithms can be easily described with
fewer data.
4. It is widely used in all fields of life and easily provides effective solutions to the
problems which have high complexity.
5. This concept is based on the set theory of mathematics, so that's why it is simple.
6. It allows users for controlling the control machines and consumer products.
7. The development time of fuzzy logic is short as compared to conventional methods.
8. Due to its flexibility, any user can easily add and delete rules in the FLS system.

Disadvantages of Fuzzy Logic

1. The run time of fuzzy logic systems is slow and takes a long time to produce outputs.
2. Users can understand it easily if they are simple.
3. The possibilities produced by the fuzzy logic system are not always accurate.
4. Many researchers give various ways for solving a given statement using this
technique which leads to ambiguity.
5. Fuzzy logics are not suitable for those problems that require high accuracy.
6. The systems of a Fuzzy logic need a lot of testing for verification and validation.
Classical and Fuzzy Set Theory
Set
A set is a term, which is a collection of unordered or ordered elements. Following are
the various examples of a set:
1. A set of all-natural numbers
2. A set of students in a class.
3. A set of all cities in a state.
4. A set of upper-case letters of the alphabet.

Types of Set:
There are following various categories of set:
• Finite
• Empty
• Infinite
• Proper
• Universal
• Subset
• Singleton
• Equivalent Set
• Disjoint Set
Classical Set
It is a type of set which collects the distinct objects in a group. The sets with the crisp
boundaries are classical sets. In any set, each single entity is called an element or
member of that set.
Mathematical Representation of Sets
Any set can be easily denoted in the following two different ways:
1. Roaster Form: This is also called as a tabular form. In this form, the set is
represented in the following way:
Set_name = { element1, element2, element3, ......, element N}
The elements in the set are enclosed within the brackets and separated by the commas.
Following are the two examples which describes the set in Roaster or Tabular form:
Example 1:
Set of Natural Numbers: N={1, 2, 3, 4, 5, 6, 7, ......,n).
Example 2:
Set of Prime Numbers less than 50: X={2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43,
47}.
2. Set Builder Form: Set Builder form defines a set with the common properties of an
element in a set. In this form, the set is represented in the following way:
A = {x:p(x)}
The following example describes the set in the builder form:
The set {2, 4, 6, 8, 10, 12, 14, 16, 18} is written as:
B = {x:2 ≤ x < 20 and (x%2) = 0}
Operations on Classical Set
Following are the various operations which are performed on the classical sets:
Union Operation
Intersection Operation
Difference Operation
Complement Operation
1. Union:

This operation is denoted by (A U B). A U B is the set of those elements which exist in
two different sets A and B. This operation combines all the elements from both the sets
and make a new set. It is also called a Logical OR operation.
It can be described as: A ∪ B = { x | x ∈ A OR x ∈ B }.
Example:
Set A = {10, 11, 12, 13}, Set B = {11, 12, 13, 14, 15}, then A ∪ B = {10, 11, 12, 13,
14, 15}
2. Intersection
This operation is denoted by (A ∩ B). A ∩ B is the set of those elements which are
common in both set A and B. It is also called a Logical AND operation.
It can be described as: A ∩ B = { x | x ∈ A AND x ∈ B }.
Example:
Set A = {10, 11, 12, 13}, Set B = {11, 12, 14} then A ∩ B = {11, 12}
3. Difference Operation
This operation is denoted by (A - B). A-B is the set of only those elements which exist
only in set A but not in set B.
It can be described as: A - B = { x | x ∈ A AND x ∉ B }.
4. Complement Operation: This operation is denoted by (A`). It is applied on a single
set. A` is the set of elements which do not exist in set A.
It can be described as: A′ = {x|x ∉ A}.

Properties of Classical Set

There are following various properties which play an essential role for finding the
solution of a fuzzy logic problem.
1. Commutative Property:
This property provides the following two states which are obtained by two finite sets A
and B:
A∪B=B∪A
A∩B=B∩A
2. Associative Property:
This property also provides the following two states but these are obtained by three
different finite sets A, B, and C:
A ∪ (B ∪ C) = (A ∪ B) ∪ C
A ∩ (B ∩ C) = (A ∩ B) ∩ C
3. Idempotency Property:
This property also provides the following two states but for a single finite set A:
A∪A=A
A∩A=A
4. Absorption Property
This property also provides the following two states for any two finite sets A and B:
A ∪ (A ∩ B) = A
A ∩ (A ∪ B) = A
5. Distributive Property:
This property also provides the following two states for any three finite sets A, B, and
C:
A∪ (B ∩ C) = (A ∪ B)∩ (A ∪ C)
A∩ (B ∪ C) = (A∩B) ∪ (A∩C)
6. Identity Property:
This property provides the following four states for any finite set A and Universal set
X:
A ∪ φ =A
A∩X=A
A∩φ=φ
A∪X=X
7. Transitive property
This property provides the following state for the finite sets A, B, and C:
If A ⊆ B ⊆ C, then A ⊆ C
8. Ivolution property
This property provides following state for any finite set A:
9. De Morgan's Law
This law gives the following rules for providing the contradiction and tautologies:
Fuzzy Set
The set theory of classical is the subset of Fuzzy set theory. Fuzzy logic is based on this
theory, which is a generalisation of the classical theory of set (i.e., crisp set) introduced
by Zadeh in 1965.
A fuzzy set is a collection of values which exist between 0 and 1. Fuzzy sets are
denoted or represented by the tilde (~) character. The sets of Fuzzy theory were
introduced in 1965 by Lofti A. Zadeh and Dieter Klaua. In the fuzzy set, the partial
membership also exists. This theory released as an extension of classical set theory.
This theory is denoted mathematically asA fuzzy set (Ã) is a pair of U and M, where U
is the Universe of discourse and M is the membership function which takes on values
in the interval [ 0, 1 ]. The universe of discourse (U) is also denoted by Ω or X.
Operations on Fuzzy Set
Given à and B are the two fuzzy sets, and X be the universe of discourse with the
following respective member functions:
The operations of Fuzzy set are as follows:
1. Union Operation: The union operation of a fuzzy set is defined by:
μA∪B(x) = max (μA(x), μB(x))
Example:
Let's suppose A is a set which contains following elements:
A = {( X1, 0.6 ), (X2, 0.2), (X3, 1), (X4, 0.4)}
And, B is a set which contains following elements:
B = {( X1, 0.1), (X2, 0.8), (X3, 0), (X4, 0.9)}
then,
AUB = {( X1, 0.6), (X2, 0.8), (X3, 1), (X4, 0.9)}
Because, according to this operation
For X1
μA∪B(X1) = max (μA(X1), μB(X1))
μA∪B(X1) = max (0.6, 0.1)
μA∪B(X1) = 0.6
For X2
μA∪B(X2) = max (μA(X2), μB(X2))
μA∪B(X2) = max (0.2, 0.8)
μA∪B(X2) = 0.8
For X3
μA∪B(X3) = max (μA(X3), μB(X3))
μA∪B(X3) = max (1, 0)
μA∪B(X3) = 1
For X4
μA∪B(X4) = max (μA(X4), μB(X4))
μA∪B(X4) = max (0.4, 0.9)
μA∪B(X4) = 0.9
2. Intersection Operation: The intersection operation of fuzzy set is defined by:
μA∩B(x) = min (μA(x), μB(x))
Example:
Let's suppose A is a set which contains following elements:
A = {( X1, 0.3 ), (X2, 0.7), (X3, 0.5), (X4, 0.1)}
And, B is a set which contains following elements:
B = {( X1, 0.8), (X2, 0.2), (X3, 0.4), (X4, 0.9)}
then,
A∩B = {( X1, 0.3), (X2, 0.2), (X3, 0.4), (X4, 0.1)}
Because, according to this operation
For X1
μA∩B(X1) = min (μA(X1), μB(X1))
μA∩B(X1) = min (0.3, 0.8)
μA∩B(X1) = 0.3
For X2
μA∩B(X2) = min (μA(X2), μB(X2))
μA∩B(X2) = min (0.7, 0.2)
μA∩B(X2) = 0.2
For X3
μA∩B(X3) = min (μA(X3), μB(X3))
μA∩B(X3) = min (0.5, 0.4)
μA∩B(X3) = 0.4
For X4
μA∩B(X4) = min (μA(X4), μB(X4))
μA∩B(X4) = min (0.1, 0.9)
μA∩B(X4) = 0.1
3. Complement Operation: The complement operation of fuzzy set is defined by:
μĀ(x) = 1-μA(x)
Example:
Let's suppose A is a set which contains following elements:
A = {( X1, 0.3 ), (X2, 0.8), (X3, 0.5), (X4, 0.1)}
then,
Ā= {( X1, 0.7 ), (X2, 0.2), (X3, 0.5), (X4, 0.9)}
Because, according to this operation
For X1
μĀ(X1) = 1-μA(X1)
μĀ(X1) = 1 - 0.3
μĀ(X1) = 0.7
For X2
μĀ(X2) = 1-μA(X2)
μĀ(X2) = 1 - 0.8
μĀ(X2) = 0.2
For X3
μĀ(X3) = 1-μA(X3)
μĀ(X3) = 1 - 0.5
μĀ(X3) = 0.5
For X4
μĀ(X4) = 1-μA(X4)
μĀ(X4) = 1 - 0.1
μĀ(X4) = 0.9

Fuzzy Reasoning:

Fuzzy reasoning involves using fuzzy logic to make decisions based on imprecise and
uncertain information. Fuzzy logic allows for the representation of vague concepts
using fuzzy sets, which assign degrees of membership to elements. Here's a detailed
explanation of fuzzy reasoning with mathematical problems and answers:

Problem: Consider a temperature control system for a greenhouse. The system needs
to decide whether to turn on the cooling system based on the current temperature and
the desired temperature range.

Solution:

1. Define Fuzzy Sets:

o Let TT be the current temperature, and DD be the desired temperature
range.
o Define fuzzy sets for TT and DD:
▪ TT can be "cold," "cool," "warm," "hot," and "very hot" with
corresponding membership functions.
▪ DD can be "low," "medium," and "high" with corresponding
membership functions.
2. Fuzzy Rules:
o Define fuzzy rules that relate TT and DD to the decision to turn on the
cooling system.
o Example rules:
▪ If TT is "hot" and DD is "low," then turn on the cooling system.
▪ If TT is "very hot" and DD is "medium," then turn on the cooling
system.
3. Fuzzy Inference:
o Use the fuzzy rules to determine the degree to which each rule applies
based on the current values of TT and DD.
o Combine the rules' outputs to determine the overall decision.
4. De-fuzzification:
o Convert the fuzzy output into a crisp decision.
o Example methods: centroid method, max membership method.

Example: Suppose the current temperature is 30°C, and the desired temperature range
is 25-28°C. Let's assume the membership functions for TT and DD have been defined.

1. Membership values for TT:

o "cool": 0.2, "warm": 0.8, "hot": 0.5, "very hot": 0.1
2. Membership values for DD:
o "low": 0.6, "medium": 0.4, "high": 0.1
3. Applying fuzzy rules:
o Rule 1: If TT is "hot" (0.5) and DD is "low" (0.6), then turn on the cooling
system.
o Rule 2: If TT is "very hot" (0.1) and DD is "medium" (0.4), then turn on
the cooling system.
4. Combining rules' outputs:
o Considering the rules' outputs, the decision to turn on the cooling system
is supported by both rules, indicating a higher likelihood of turning on the
cooling system.
5. Defuzzification:
o Using the centroid method, the crisp decision could be to turn on the
cooling system at a certain level to bring the temperature within the
desired range.

Fuzzy reasoning allows for more nuanced decision-making in situations where

traditional logic may not be suitable due to uncertainty or imprecision in the input data.

Fuzzy Inference System

Fuzzy Inference System is the key unit of a fuzzy logic system having decision making
as its primary work. It uses the “IF…THEN” rules along with connectors “OR” or
“AND” for drawing essential decision rules.

Characteristics of Fuzzy Inference System

Following are some characteristics of FIS −

• The output from FIS is always a fuzzy set irrespective of its input which can be
fuzzy or crisp.
• It is necessary to have fuzzy output when it is used as a controller.
• A defuzzification unit would be there with FIS to convert fuzzy variables into crisp
variables.

Functional Blocks of FIS

The following five functional blocks will help you understand the construction of FIS −

• Rule Base − It contains fuzzy IF-THEN rules.

• Database − It defines the membership functions of fuzzy sets used in fuzzy rules.
• Decision-making Unit − It performs operation on rules.
• Fuzzification Interface Unit − It converts the crisp quantities into fuzzy
quantities.
• Defuzzification Interface Unit − It converts the fuzzy quantities into crisp
quantities. Following is a block diagram of fuzzy interference system.

Working of FIS

The working of the FIS consists of the following steps −

• A fuzzification unit supports the application of numerous fuzzification methods,

and converts the crisp input into fuzzy input.
• A knowledge base - collection of rule base and database is formed upon the
conversion of crisp input into fuzzy input.
• The defuzzification unit fuzzy input is finally converted into crisp output.

Methods of FIS

Let us now discuss the different methods of FIS. Following are the two important
methods of FIS, having different consequent of fuzzy rules −

• Mamdani Fuzzy Inference System

• Takagi-Sugeno Fuzzy Model (TS Method)
Mamdani Fuzzy Inference System

This system was proposed in 1975 by Ebhasim Mamdani. Basically, it was anticipated
to control a steam engine and boiler combination by synthesizing a set of fuzzy rules
obtained from people working on the system.

Steps for Computing the Output

Following steps need to be followed to compute the output from this FIS −

• Step 1 − Set of fuzzy rules need to be determined in this step.

• Step 2 − In this step, by using input membership function, the input would be
made fuzzy.
• Step 3 − Now establish the rule strength by combining the fuzzified inputs
according to fuzzy rules.
• Step 4 − In this step, determine the consequent of rule by combining the rule
strength and the output membership function.
• Step 5 − For getting output distribution combine all the consequents.
• Step 6 − Finally, a defuzzified output distribution is obtained.

Following is a block diagram of Mamdani Fuzzy Interface System.

Takagi-Sugeno Fuzzy Model (TS Method)

This model was proposed by Takagi, Sugeno and Kang in 1985. Format of this rule is
given as −

IF x is A and y is B THEN Z = f(x,y)

Here, AB are fuzzy sets in antecedents and z = f(x,y) is a crisp function in the
consequent.

Fuzzy Inference Process

The fuzzy inference process under Takagi-Sugeno Fuzzy Model (TS Method) works in
the following way −

• Step 1: Fuzzifying the inputs − Here, the inputs of the system are made fuzzy.
• Step 2: Applying the fuzzy operator − In this step, the fuzzy operators must be
applied to get the output.

Rule Format of the Sugeno Form

The rule format of Sugeno form is given by −

if 7 = x and 9 = y then output is z = ax+by+c

Comparison between the two methods

Let us now understand the comparison between the Mamdani System and the Sugeno
Model.

• Output Membership Function − The main difference between them is on the

basis of output membership function. The Sugeno output membership functions
are either linear or constant.
• Aggregation and Defuzzification Procedure − The difference between them
also lies in the consequence of fuzzy rules and due to the same their aggregation
and defuzzification procedure also differs.
• Mathematical Rules − More mathematical rules exist for the Sugeno rule than
the Mamdani rule.
• Adjustable Parameters − The Sugeno controller has more adjustable parameters
than the Mamdani controller.
Deep learning architecture for Fuzzy logic or Fuzzy Neural Network
A fuzzy neural network or neuro-fuzzy system is a learning machine that finds the
parameters of a fuzzy system (i.e., fuzzy sets, fuzzy rules) by exploiting approximation
techniques from neural networks.

Combining fuzzy systems with neural networks

Both neural networks and fuzzy systems have some things in common. They can be used
for solving a problem (e.g. pattern recognition, regression or density estimation) if there
does not exist any mathematical model of the given problem. They solely do have certain
disadvantages and advantages which almost completely disappear by combining both
concepts.

Neural networks can only come into play if the problem is expressed by a sufficient
amount of observed examples. These observations are used to train the black box. On the
one hand no prior knowledge about the problem needs to be given. On the other hand,
however, it is not straightforward to extract comprehensible rules from the neural
network's structure.

On the contrary, a fuzzy system demands linguistic rules instead of learning examples as
prior knowledge. Furthermore the input and output variables have to be described
linguistically. If the knowledge is incomplete, wrong or contradictory, then the fuzzy
system must be tuned. Since there is not any formal approach for it, the tuning is
performed in a heuristic way. This is usually very time consuming and error-prone.

Table 1: Comparison of neural control and fuzzy control

Neural Networks Fuzzy Systems
no mathematical model necessary no mathematical model necessary
learning from scratch apriori knowledge essential
several learning algorithms not capable to learn
black-box behavior simple interpretation and implementation
It is desirable for fuzzy systems to have an automatic adaption procedure which is
comparable to neural networks. As it can be seen in Table 1, combining both approaches
should unite advantages and exclude disadvantages.

Characteristics

Compared to a common neural network, connection weights and propagation and

activation functions of fuzzy neural networks differ a lot. Although there are many
different approaches to model a fuzzy neural network (Buckley and Hayashi, 1994, 1995;
Nauck and Kruse, 1996), most of them agree on certain characteristics such as the
following:

Figure 1: The architecture of a neuro-fuzzy system

• A neuro-fuzzy system based on an underlying fuzzy system is trained by means of

a data-driven learning method derived from neural network theory. This heuristic
only takes into account local information to cause local changes in the fundamental
fuzzy system.
• It can be represented as a set of fuzzy rules at any time of the learning process, i.e.,
before, during and after.
o Thus the system might be initialized with or without prior knowledge in
terms of fuzzy rules.
• The learning procedure is constrained to ensure the semantic properties of the
underlying fuzzy system.
• A neuro-fuzzy system approximates a n-dimensional unknown function which is
partly represented by training examples.
o Fuzzy rules can thus be interpreted as vague prototypes of the training data.
• A neuro-fuzzy system is represented as special three-layer feedforward neural
network as it is shown in Figure 1.
o The first layer corresponds to the input variables.
o The second layer symbolizes the fuzzy rules.
o The third layer represents the output variables.
o The fuzzy sets are converted as (fuzzy) connection weights.
o Some approaches also use five layers where the fuzzy sets are encoded in
the units of the second and fourth layer, respectively. However, these
models can be transformed into a three-layer architecture.

One can basically distinguish between three different kinds of fuzzy neural networks,
i.e., cooperative, concurrent and hybrid FNNs (Nauck et al., 1997).

Cooperative Fuzzy Neural Network

Figure 2: Different cooperative fuzzy neural networks

In the case of cooperative neural fuzzy systems, both artificial neural network and fuzzy
system work independently from each other. The ANN tries to learn the parameters from
the fuzzy system. This can be either performed offline or online while the fuzzy system
is applied. Figure 2 depicts four different kinds of cooperative fuzzy neural networks.
The upper left fuzzy neural network learns fuzzy set from given training data. This is
usually performed by fitting membership functions with a neural network. The fuzzy sets
are then determined offline. They are then utilized to form the fuzzy system by fuzzy
rules that are given (not learned) as well.

The upper right neuro-fuzzy system determines fuzzy rules from training data by a neural
network. Here as well, the neural networks learns offline before the fuzzy system is
initialized. The rule learning usually done by clustering on self-organizing feature maps
(Bezdek et al., 1992; Vuorimaa, 1994). It is also possible to apply fuzzy clustering
methods to obtain rules.

In the lower left neuro-fuzzy model, the system learns all membership function
parameters online, i.e., while the fuzzy system is applied. Thus initially fuzzy rules and
membership functions must be defined beforehand. Moreover, the error has to be
measured in order to improve and guide the learning step.

The lower right one determines rule weights for all fuzzy rules by a neural network. This
can be done online and offline. A rule weight is interpreted as the influence of a rule
(Kosko, 1992). They are multiplied with the rule output. In (Nauck et al., 1997) the
authors argue that the semantics of rule weights are not clearly defined. They could be
replaced by modified membership functions. However, this could destroy the
interpretation of fuzzy sets. Moreover, identical linguistic values might be represented
differently in dissimilar rules.

Hybrid Fuzzy Neural Network

Figure 3: A hybrid fuzzy neural network

Hybrid neuro-fuzzy systems are homogeneous and usually resemble neural networks.
Here, the fuzzy system is interpreted as special kind of neural network. The advantage
of such hybrid NFS is its architecture since both fuzzy system and neural network do not
have to communicate any more with each other. They are one fully fused entity. These
systems can learn online and offline. Figure 3 shows such a hybrid FNN.

The rule base of a fuzzy system is interpreted as a neural network. Fuzzy sets can be
regarded as weights whereas the input and output variables and the rules are modeled as
neurons. Neurons can be included or deleted in the learning step. Finally, the neurons of
the network represent the fuzzy knowledge base. Obviously, the major drawbacks of both
underlying systems are thus overcome.

In order to build a fuzzy controller, membership functions which express the linguistic
terms of the inference rules have to be defined. In fuzzy set theory, there does not exist
any formal approach to define these functions. Any shape (e.g., triangular, Gaussian) can
be considered as membership function with an arbitrary set of parameters. Thus the
optimization of these functions in terms of generalizing the data is very important for
fuzzy systems. Neural networks can be used to solve this problem.

By fixing a distinct shape of the membership functions, say triangular, the neural network
must optimize their parameters by gradient descent (Nomura et al., 1992). Thus, aside
information about the shape of the membership functions, training data must be available
as well.

Another approach (Hayashi et al., 1992) is to group the training data

{(xi,yi)|xi∈X,yi∈Y,i=1,2,…,l} into M clusters. Every cluster represents a rule Rm where
m=1,2,…,M . Hence these rules are not defined linguistically but rather by crisp data
points x=(x1,x2,…,xn) . Thus a neural network with n input units, hidden layers and M
output units might be applied to train on the pre-defined clusters. For testing, an arbitrary
pattern x is presented to the trained neural network. Every output unit m will return a
degree to which extend x may fit to the antecedent of rule Rm .

To guarantee the characteristics of a fuzzy system, the learning algorithm must enforce
the following mandatory constraints:
• Fuzzy sets must stay normal and convex.
• Fuzzy sets must not exchange their relative positions (they must not pass each
other).
• Fuzzy sets must always overlap.

Additionally there do exist some optional constraints like the following:

• Fuzzy sets must stay symmetric.

• The membership degrees must sum up to 1.

An important hybrid fuzzy neural network has been introduced in (Berenji, 1992). The
ARIC (approximate reasoning-based intelligent control) is presented as a neural network
where a prior defined rule base is tuned by updating the network's prediction. Thus the
advantages of fuzzy systems and neural networks are easily combined as presented in
Table 1.

The ARIC is represented by two feed-forward neural networks, the action-state

evaluation network (AEN) and the action selection network (ASN). The ASN is a
multilayer neural network representation of a fuzzy system. It then again consists of two
separate. The first one represents the fuzzy inference and the second one computes a
confidence measure based on the current and next system state. Both parts are eventually
combined to the ASN's output.

As it is shown in Figure 1, the first layer represents the rule antecedents, whereas the
second layer corresponds to the implemented fuzzy rules and the third layer symbolized
the system action. The network flow is at follows. In the first layer the system variables
are fuzzified. In the next step these membership values are multiplied by the attached
weights of the connections between the first and second layer. In the latter layer, every
rule's input corresponds to the minimum of its input connections.

A rule's conclusion is installed as membership function. This function maps the inverse
rule input value. Its output values is then multiplied by the weights of the connections
between second and third layer. The final output value is eventually computed by the
weighted average of all rules' conclusions.

The AEN (which is as three-layer feed-forward neural network as well) aims to forecast
the system behavior. The hidden layer obtains as input both the system state and an error
signal from the underlying system. The output of the networks shall represent the
prediction of the next reinforcement which depends on the weights and the system state.
The weights are changed by a reinforcement procedure which takes into consideration
the outputs of both networks ASN and AEN, respectively. ARIC was successfully
applied to the cart-pole balancing problem.

Whereas the ARIC model can be easily interpret as a set of fuzzy-if-then rules, the ASN
network to adjust the weights is rather difficult to understand. It is a working neural
network architecture that utilizes aspects of fuzzy systems. However, a semantic
interpretation of some learning steps is not possible.

Berenji and Khedkar (1992) introduced an improvement of the their former approach
named GARIC (generalized ARIC). This idea does not suffer from different
interpretations of the linguistic values anymore by refraining from weighted connections
in the ASN. Instead the fuzzy sets are represented as nodes in the network. Moreover the
learning procedure changes parameters of these nodes and thus the shape of the
membership functions. GARIC is also able to use any kind of membership functions in
the conclusion since a different defuzzifier and a differentiable soft-minimum function
are used.