Reference taken from some website

https://www.javatpoint.com/single-layer-perceptron-in-tensorflow
https://www.datasciencecentral.com/learning-rules-in-neural-network/
https://towardsdatascience.com/understanding-backpropagation-algorithm-
7bb3aa2f95fd
https://www.slideshare.net/AMITKUMAR4132/fuzzy-set-theory
https://www.educba.com/what-is-genetic-algorithm/

Q : Compare and contrast Biological Neuron and Artificial Neuron with a

suitable diagram
Q : Discuss about the name of a network that includes backward links from a
given output to its inputs along with the hidden layers?
Ans : Recurrent neural network

Define Soft Computing. How is it different from conventional computing?

Write down the applications of genetic algorithm.
Why do we use bias function in neural network?
There are a Function which takes some input which we say Antecedent and
giving input “x” and we are doing computing in Function x so it’ll give some
output which stores in Y which we can say consequent
Here this is single function f(x)

Here function x is doing computing process which we can say this function F
Formal Method / Algorithm / Mapping Function
Here in picture we can say that we get desired output in consequent which is
done by control Action
Features of Soft Computing
• Precise Solution
• Unambiguous & Accurate
• Mathematical Model
Precise Solution
If we use computing and does problem solving then it should be precise
manner
Unambiguous & Accurate
As we have mentioned computing unit which transforms data from function
that is nothing but which is algorithm
And algorithms process the data by control action
So control action should be unambiguous and algorithm should not be two
meaning
Mathematical
This is also algorithm to solve the problem
Soft Computing Technique
Computation is a process of converting the input of one form to some other
desired output form using certain control actions. According to the concept of
computation, the input is called an antecedent and the output is called the
consequent. A mapping function converts the input of one form to another
form of desired output using certain control actions. The computing concept is
mainly applicable to computer science engineering. There are two types of
computing, hard computing, and soft computing.
Hard Computing
• Precise Results
Hard computing is a process in which we program the computer to solve
certain problems using mathematical algorithms that already exist, which
provides a precise output value. One of the fundamental examples of hard
computing is a numerical problem.
This will be precise manner because in Hard computing allows mathematical
operation to solve algorithm which exists on the method
• Control Action
• Unambiguous Data
Control action allows the mathematical operation which performs on
algorithm

So whereas mathematical operation exists there is need for hard

computing

Quick sort
Bubble Sort
Merge sort
Insertion Sort
 Heap sort
Selection
Computational Geometry Problem
In mathematics Graph x and y coordinate

In come cases there are uncertainty the output

Imprecision

Dynamic -> The result may change in the output because what algorithm
size is changes that is not adaptive

Uncertainty

When programmer runs any program then it is not the same data as user
wants because user also apply implementation in different manner

Low solution cost

Because lack of service and when we use the hard computing that is High
Solution Cost due precise and unambiguous method

Fuzzy Logic : Lets take example of Doctor and Patient

its super set of Boolean logic means complete true or false Yes/No
But in fuzzy logic defines uncertanity
Patients shares his problem and the doctor considers the technique of cure
how does it match the exact disease
But in some case any symptom that matches 60% to 70% for any disease on
that basis doctor prescribes medicine while 30% to 40% in uncertainity
Hence in Fuzzy logic relates the uncertain data
Neural Network
A, a, a, a all is A by the knowledge base which learn any method
So soft computing is replacing Neural Network
Evolutionary / Generic Computing - > Decision base computing carried out
Who will win any match from the last inning cricket match
What is Soft Computing?
Soft computing is an approach where we compute solutions to the
existing complex problems,
where output results are imprecise or fuzzy in nature, one of the most
important features of soft computing is it should be adaptive so that any
change in environment does not affect the present process. The following are
the characteristics of soft computing.
• It does not require any mathematical modeling to solve any given
problem
• It gives different solutions when we solve a problem of one input from
time to time
• Uses some biologically inspired methodologies such as genetics,
evolution, particles swarming, the human nervous system, etc.
• Adaptive in nature.
There are three types of soft computing techniques which include the
following.
• Fuzzy Logic
• Artificial Neural Network
• Genetic algorithm

Example of Fuzzy Logic as comparing to Boolean Logic

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.
Fuzzy Logic
The fuzzy logic algorithm is used to solve the models which are based on
logical reasoning like imprecise and vague. It was introduced by Latzi A.
Zadeh in 1965. Who was a mathematician Fuzzy logic fuzzy logic provides
stipulated truth value with the closed interval [0,1]. Where 0 = false value, 1=
true value.
Fuzzy Logic is a super set of Boolean Logic as we have a conventional
method to represent it i.e, 0 and 1 means Yes or No, True or false
But Fuzzy logic represent uncertainty means it works on the degree and
Boolean logic will not work
Means we are representing degree the speed of
An example of a robot that wants to move from one place to another within a
short time where there are many obstacles on the way. Now the question
arises is that how the robot can calculate its movement to reach the
destination point, without colliding to any obstacle. These types of problems
have uncertainty problem which can be solved using fuzzy logic.

Example 2

string1 = "xyz" and string2 = "xyw"

Problem 1
Are string1 and string2 same?
No, the solution is simply No. It does not require any algorithm to analyse
this.
Let's modify the problem a bit.

Problem 2
How much string1 and string2 are same?
Solution
Through conventional programming, either the answer is Yes or No. But
these strings might be 80% similar according to soft computing.
You have noticed that soft computing gave us the approximate solution.
 Fuzzy Set Crisp Set

It is prescribed by vague or It is defined by precise and

Basic
ambiguous properties. specific characteristics.

It is a set of components with It is a set of objects that have

Definition different membership the same countability and
degrees in the set. finiteness qualities.

It is commonly utilized in It is commonly utilized in digital

Applications
fuzzy controllers. design.

It shows incomplete It shows the complete

Membership
membership. membership.

It follows the infinite-valued

Logic It follows the bi-valued logic.
logic.

It specifies a number
It specifies the value as either 0
Value between 0 and 1, which
or 1.
includes both 0 and 1.

It defines the degree to which It is also referred to as a classical

Degree
anything is true. set.
Solution

Artificial Neural Network

Neural networks were developed in the 1950s, which helped soft computing
to solve real-world problems, which a computer cannot do itself. We all know
that a human brain can easily describe real-world conditions, but a computer
cannot

This works on LLM (Large Level Model)

Some Basic Components of ANN

 • Neurons (Nodes)

• Layers

• Connections

• Activation Function

Neurons

Neurons is like Human Brain what we think and gives the output Neuron Is Just
Like Store the data

Data is stored and it will calculating the data now it is creating Layer when its
processing the data

Layers Input data

Data Stores Hiiden layer data is breaking and associated

Layers are multiple Layers

Connection

The above picture is a Neuron Network

Activation Function

Output becomes sigmear data but we want that our value should come in None
Linear Data between (0 & 1) then apply Activation function

Over all the in above picture we are doing input as training data and it gives
relevant data as we want because in every layer Data is relating as desired
output

X1.w1 + X2.w2 + b (Biased Value)

• It is a connectionist modelling and parallel distributed network. There are

of two types ANN (Artificial Neural Network) and BNN (Biological Neural
 Network). A neural network that processes a single element is known as
a unit. The components of the unit are, input, weight, processing element,
and output. It is similar to our human neural system. The main advantage
is that they solve the problems in parallel, artificial neural networks use
electrical signals to communicate. But the main disadvantage is that they
are not fault-tolerant that is if anyone of artificial neurons gets damaged
it will not function anymore.

Large Language Model (LLM) used Artificial Intelligence

An example of a handwritten character, where a character is written in

Hindi by many people, they may write the same character but in a
different form. As shown below, whichever way they write we can
understand the character, because one already knows how the
character looks like. This concept can be compared to our neural
network system.

Evolutionary / Genetic Algorithm in Soft Computing

The genetic algorithm was introduced by Prof. John Holland in 1965. It is
used to solve problems based on principles of natural selection, that come
under evolutionary algorithm. They are usually used for optimization
problems like maximization and minimization of objective functions,
which are of two types of an ant colony and swarm particle. It follows
biological processes like genetics and evolution.
This is Decision Based Computation Analysis
Who win the next IPL Match
CSK
KRK
RCB

Some characteristics of Soft computing

• Soft computing provides an approximate but precise solution for real-
life problems.
• The algorithms of soft computing are adaptive, so the current process
is not affected by any kind of change in the environment.
• The concept of soft computing is based on learning from experimental
data. It means that soft computing does not require any mathematical
model to solve the problem.
• Soft computing helps users to solve real-world problems by providing
approximate results that conventional and analytical models cannot
solve.
• It is based on Fuzzy logic, genetic algorithms, machine learning, ANN,
and expert systems.

Example
Soft computing deals with the approximation model. You will understand
with the help of examples of how it deals with the approximation model.

Let's consider a problem that actually does not have any solution via
traditional computing, but soft computing gives the approximate solution.

Functions of the Genetic Algorithm

The genetic algorithm can solve the problems which cannot be solved in
real-time also known as the NP-Hard pLLMroblem. The complicated
problems which cannot be solved mathematically can be easily solved by
applying the genetic algorithm.
A simple way of understanding this algorithm is by considering the
following example of a person who wants to invest some money in the
bank, we know there are different banks available with different schemes
and policies. Its individual interest how much amount to be invested in the
bank, so that he can get maximum profit. There are certain criteria for the
person that is, how he can invest and how can he get profited by investing
in the bank. These criteria can be overcome by the “Evolutional
Computing” algorithm like genetic computing.

Difference Between Hard Computing and Soft Computing

The difference between hard computing and soft computing are as follows

Hard Computing Soft Computing

• The analytical model required by • It is based on uncertainty, partial
hard computing must be precisely truth tolerant of imprecision and
represented approximation.
• Computation time is more • Computation time is less
• It depends on binary logic, numerical • Based on approximation and
systems, crisp software. dispositional.
• Sequential computation • Parallel computation
• Gives exact output • Gives appropriate output
Examples: Traditional methods of • Example: Neural networks like
computing using our personal Adaline, Madaline, ART networks,
computer. etc.

Advantages
The benefits of soft computing are
• The simple mathematical calculation is performed
• Good efficiency
• Applicable in real-time
• Based on human reasoning.
Disadvantages
The disadvantages of soft computing are
• It gives an approximate output value
• If a small error occurs the entire system stops working, to overcome its
entire system must be corrected from the beginning, which is time
taking process.
Applications
The following are the applications of soft computing
• Controls motors like induction motor, DC servo motor automatically
• Power plants can be controlled using an intelligent control system
• In image processing, the given input can be of any form, either image or
video which be manipulated using soft computing to get an exact
duplicate of the original image or video.
• In biomedical applications where it is closely related to biology and
medicine, soft computing techniques can be used to solve biomedical
problems like diagnosis, monitoring, treatment, and therapy.
Computing is a technique used to convert particular input using control
action to the desired output. There are two types of computing
techniques hard computing and soft computing. Here in our article, we
are mainly focusing on soft computing, its techniques like fuzzy logic,
artificial neural network, genetic algorithm, comparison between hard
computing and soft computing, soft computing techniques, applications,
and advantages. Here is the question “How are soft computing is
applicable in the medical field?”

Evolution of Mathematic is Fuzzy Logic

 Fuzzy Logic

Fuzzy Logic

+
+

There are three types of soft computing techniques which include the
following.
 • Fuzzy Logic
• Artificial Neural Network
• Genetic algorithm
Example of Fuzzy Logic as comparing to Boolean Logic

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
Artificial Neural Network

Neural networks were developed in the 1950s, which helped soft computing
to solve real-world problems, which a computer cannot do itself. We all know
that a human brain can easily describe real-world conditions, but a computer
cannot

Here Neural Means Neuron and Network which is connected in them self

Like Internet which does sharing of Data

In Humans there are connected high number of neuron
connected in each and other they think and Transfer the data in
information and sharing of data
Neurons

Neurons is like Human Brain what we think and gives the output Neuron Is Just
Like Store the data

Fuzzy Set

Universal Set and Our Set

Membership Belong

U (meun)

U = {1, 2, 3, 4, 5 }

S = {1, 2}

={(1,1) (2,1),(3,0),(4,0),(,5,0)}

x,u

Here 1 is with 1 which is membership values =1

Similiarly Here 2 is with 1 which is membership values =1

Here 3, 4 and 5 are with 0 which is membership values =0

Fuzzy set theory, introduced by Lotfi A. Zadeh, redefines the conventional notion
of set theory by accommodating the granularity of membership within a set.
Unlike classical set theory, which employs binary membership functions, fuzzy sets
allow for a continuum of membership grades, thereby enabling the representation
of partial truths.
Fuzzy Logic

Fuzzy logic is a mathematical method for reasoning that's approximate and

resembles human decision-making. It's based on the idea that real-world
information is often vague and partially true, and that there are many
possibilities between yes = 1 and no = 0
It is super set of Boolean logic means in Boolean Logic we represent [0 and 1]

It represents uncertainty it represent the degree in membership function

Fuzzy set can have a progressive transition among many degrees of

membership.

Means values will come between in 0 and 1 like 0.2, 0.3, 0.1

Speed Transition Apply smart Agent Result Fuzzy logic

Slow : 30.4, 39.9 -> Acceleration Oscillate Flexibility

Fast : 40, 40.1, 41 Break Oscillate Flexibility

In this pic we can represent
flexible way via degree as
given picture

Boolean logic Lets say we have universal set U =

to find out whether this sub {1,2,3,4,5} and we have subset
set belong to U means S value which value is S = {1,2}
belongs to Universal set U µ->membership
{(1,1),(2,1),(3,0),(4,0),(5,0)}

Check the degree of fastness

of car through fuzzy logic
Represents the belonging of a
member function of a crisp set
to fuzzy set

0, if speed(x) <=40 x=30 (30,0)

1, if speed(x) >=50 x=60 (60,1)
(speed(x)-40)/10, 42-40/10 =2/10=1/5 =0.2 up to degree
if 40<speed(x)<50 48-40/10 = 8/10=4/5=0.8

Membership function
And we have to find out whether this sub set belong to U means S
value belongs to Universal set U

={1,1} 1 contains in U set Yes then 1 here U=1->x and 1 Yes 1 -> µ
(meun) then its membership value is 1

Next ={(1,1), (2,1)}

Again check from Universal set

{(1,1),(2,1),(3,0),(4,0),(5,0)}

Hera membership values lies between 0 and 1 in Boolean but fuzzy

shows the belongingness of a membership

It denotes Degree

Speed (x) -50/10 :

Fuzzy Set

A fuzzy set is a mathematical technique that uses a membership function

to describe the degree of membership of an element in a set
Universal Set X ->x

The fuzzy set A defined on X is a collection of ordered pair

A = {(x, µA (x)}

Where µA (x): X-> [0,1] is called membership function

It represents of degree of membership function of element x with set A

which is value between 0 and 1

we consider Universal Set X and consider fuzzy set A i.e, closed to 3

X = {1,2,3,4,5,6,7}

A->number closed to 3

Given x to be a universe of discourse and A and B to be fuzzy sets µA(x) and

µB(x) are their respective membership function

Union : Max {µA(x),µB(x)}, x Ɛ U

or operation in Boolean

x µ
A = {(10, 0.2), (20, 0.4), (25,0.7), (30,0.9), (40,1)}
B = {(10,0.4), (20, 0.1), (25,0.9),(30, 0.2), (40,0.6)}

A U B = {(10,0.4),(20,0.4), (25,0.9), (30,0.9), (40,1)}

Example in python :

key=[]
k=0
for i in range(5):
k+=10
key.append(k)
print("Set Value : ",key)

i=0
µc=[]
x=[0.8,0.3,0.4,0.6,0.2]
y=[0.5,0.7,0.9,0.1,0.3]
while i < len(x):
if x[i] > y[i]:
µc.append(x[i])
elif y[i] > x[i]:
µc.append(y[i])
i+=1

print(µc)
bold=dict(zip(key,µc))
print(bold)

Output :
Set Value : [10, 20, 30, 40, 50]
[0.8, 0.7, 0.9, 0.6, 0.3]
{10: 0.8, 20: 0.7, 30: 0.9, 40: 0.6, 50: 0.3}

Intersection : Min {µA(x),µB(x)}, x Ɛ U

And operation in Boolean

A Ո B = {(10,0.2), (20,0.1), (25,0.7), (30, 0.2), (40,0.6)}

some time we need to find out µ value ex: µ=x/(x+2)

Complement : µA̅(x) = [1 - µ(x)] x Ɛ U

Let A = {(10, 0.2), (20, 0.4), (25,0.7), (30,0.9), (40,1)}

Solution :

µA̅(x) = (10, 0.8), (20,0.6), (25, 0.3), (30,0.1), (40, 0)

Example of complement in python
the uniform() method returns a random floating number between the
two specified numbers.

import random
num=int(input("Enter the size of Element in set.."))
if num > 5:
print("Enter size 5 ")
else:
b=[]
h=[]
j=0
for n in range(num):
t=random.uniform(0,1)
g=round(t,1)
b.append(g)
print("membership : =",b)

if num > 5:
print("Enter size 5")
else:
c=[]
k=0
for m in range(num):
k+=10
c.append(k)
print("Set Value : ",c)
print("\n")
i=0
µA=list()
while i < len(c):
t=1-b[i]
s=round(t,2)
µA.append(s)
i+=1
print(µA)
comp=dict(zip(c,µA))
print("Fuzzy Complement ")
print(comp)

Output :
Enter the size of Element in set..4
membership : = [0.9, 0.8, 0.8, 0.2]
Set Value : [10, 20, 30, 40]

[0.1, 0.2, 0.2, 0.8]

Fuzzy Complement
{10: 0.1, 20: 0.2, 30: 0.2, 40: 0.8}

Bold Union : µ A ⊕ B = Min [1, µA (x) + µB (x)]

Here first Add both A and B µ Value and arrange Minimum Value also check
from 1 if the µ vaalue is greater than 1 then place 1 µ value [0 – 1]

ex : if µ=1 and 1.6 then min. value of µ=1

Min(1,0.6) -> 0.6, Min(1,0.5) -> 0.5, Min(1,1.6) -> 1

(10,0.6), (20,0.5), (25,1), (30,1), (40,1)

x µ
A = {(10, 0.2), (20, 0.4), (25,0.7), (30,0.9), (40,1)}
B = {(10,0.4), (20, 0.1), (25,0.9),(30, 0.2), (40,0.6)}

Bold Intersection : µA o B (x) = Max [0,µA (x) + µB (x) – 1]

µA o B (x) = (0.2 + 0.4) – 1

= (0.6) – 1 = – 0.4 compare to 0 which is greater than – 0.4 so it will

be maximum value 0

= (10, 0)

= (20, 0.4) and (20, 0.1) again (0.4+0.1) – 1 = (0.5) – 1 = – 0.5

again – 0.5 it will compare to 0 then maximum is 0 i.e, (20,0)

again check (25,0.7) and (25,0.9) = (0.7+0.9) = (1.6) – 1 = 0.6

so 0.6 it will compare to 0 then maximum is 0.6 i.e, (25, 0.6)

= (30,0.9) and (30,0.2) again (0.9+0.2) = 1.1 – 1 = 0.1 and compare to 0 then
maximum value is 0.1 (30,0.1)

= (40,1) and (40,0.6) again (1+0.6)=1.6 – 1 = 0.6 and compare to 0 then

maximum value is 0.6 (40,0.6)

= {(10,0), (20,0), (25,0.6), (30,0.1), (40,0.6)}

Equality

A=B if µA (x) = µB (x)Ɐ x Ɛ S

A = {(10, 0.2), (20, 0.4), (25,0.7), (30,0.9), (40,1)}

B = {(10,0.4), (20, 0.1), (25,0.9),(30, 0.2), (40,0.6)}

here A = (10, 0.2) µA (x) = 0.2

here B = (10,0.4) µB (x) = 0.4

Both are not equal

Product

µA (x) . µB (x) = (10, 0.2) . (10, 0.4) = (10, 0.08)

Ex : Dictionary :

key=["A", "B", "C", "D"]

value=[0.2,0.8,0.1,0.6]

print(dict(zip(key,value)))

Ex :

dicts = {}

keys = range(4)

values =[0.2,0.8,0.1,0.6]

for i in keys:

dicts[i] = values[i]

print(dicts)

Output :

{0: 0.2, 1: 0.8, 2: 0.1, 3: 0.6}

Crisp Set or Classical set

Classical set theory also termed as Crisp

Theory of crisp set had its root of Boolean logic

Crisp set is also called classical set

Crisp sets are classical sets defined in boolean logic

In crisp set known as function known as characteristic function

Crisp set defines the value is either 0 or 1.

Roster notation or Tabular Form

A = {a, e, i, o, u}

B = (1,3,5,7,9}

 The elements in roster form can be in any order (they don't need to be in
ascending/descending order).
 The elements should not be repeated in set roster notation.

One of the limitations of roster notation is that we cannot represent a large number
of data in roster form.

if we want to represent the first 100 or 200 natural number in a set B the limitation
we can be overcome by representing data with the help of a dotted line. then it is
necessary to represent large data i.e, odd number between 1 to 100
{1,3,5,7,……….,199}

Builder Notation

Such that’ is expressed by the symbols ‘|’ (a vertical bar separator) or ‘:’ (colon).

the set {a,e,i,o,u} is written as A = {x | x is a vowel alphabet element }

the set {1,3,5,7,9} is written as A = {x | x ∈ ℕ, 1 < x < 10, x is odd} and is read as
"set A is the set of all ‘x’ such that ‘x’ is a odd number between 1 and 10."

The set {1,3,5,7,9} is writing as B = {x | x < 10 and (x%2 0) }

A = {x | x ∈ ℕ, 5 < x < 10} and is read as "set A is the set of all ‘x’ such that ‘x’ is
a natural number between 5 and 10."

Real Number :-> union of rational number and irrational number, these number can
be expressed on a number line

A set of real numbers less than 8 is written in set builder notation as follows:
Types of Set

Singleton Set

A set which contains a single element is called a singleton set.

Example: There is only one apple in a basket of grapes.

Check whether the given sets are singleton sets or not?

S = {10}

Example : The given set is A = {1, 3, 5, 7, 11}

Therefore the five singleton sets which are subsets of the given set A is {1}, {3}, {5},
{7}, {11}.

Example :

Set Q = {|x|: x2 = 16}

Disjoint Set

Disjoint

In order to find if two sets are disjoint sets, we need to perform the intersection of
sets operation on these two sets. The condition for any given sets to be disjoint can
be given as A ∩ B = 𝛟
P = {1, 2}, Q = {2, 3} and R = {5, 3}.
P ∩ Q = {1, 2} ∩ {2, 3}

P ∩ Q = {1,2} ∩ {2,3}= {2}

And,

Q ∩ R = {2, 3} ∩ {5, 3} = {3}

And then the last pair,

P ∩ R = {1, 2} ∩ {5, 3}

P∩R=𝛟
Example in Python
p = {1, 2}
q = {2, 3}
r = {5, 3}

res = p.intersection(q, r)
print(res)

Output :
set()

Example :
p = {1, 2,1,4,6}
q = {2,3,4,6}
r = {5,4,3}
s = {4,9,12,16}
m = p.intersection(q)
print(m)
n = q.intersection(r)
print(n)
t = p.intersection(q,r,s) // p ∩ q ∩ r ∩ s
print(t)

Output :
{4}

So, among the given sets, P and R can be considered as disjoint sets.
Disjoint sets:
Two sets A and B are said to be disjoint, if they do not have any common element
in them, i.e. A ∩ B = { }. For sets

A = Set of even numbers = {2, 4, 6, 8} and

B = Set of odd numbers = {1, 3, 5, 7}

A ∩ B = { }, so here A and B are disjoint sets

Finite set

Finite sets can be easily represented in roster notation form. For example, the set
of vowels in English alphabets, Set A = {a, e, i, o, u}

The finite set is countable and contains a finite number of elements

A set which contains a definite number of element called a finite set

Example – S = {x | 1< x < 10 and (x%2 0)}

Here x is range of 1 to 10 and all odd no. 1,3,5,7,9

Infinite set

A set which contains infinite number of element is called an infinite set

The elements of infinite sets are endless, that is, infinite.

Example – S = {x | x ∈ N and x > 10}

Here x such that x is a integer and x is greater than 10 i.e, 11,12,13,………. infinite
no.

Example : The set of whole numbers, W = {0, 1, 2, 3, ……..} is an infinite

Empty Set or Null Set

An empty set contains no element it is denoted by ɸ

Example – S = {x | x ∈ N and 7 < x < 8} = ɸ

Here x is a integer and x is greater than 7 and less than 8, so there is no number
which is greater than 7, and less than 8 means it contains 0 no. element

so this is called empty set here

Subset

A set X is a subset of Y (written as X ⊆ Y) if every element of X is an element of set

Y

Here X is subset of Y and Y is Superset of X

Example Y = {1,2,3,4,5,6} and X = {1,2} so we can write X ⊆ Y

Example A = {0,1,2} and B = {0,1,2} so here all element of A is in B

A ⊆ B and we can also define B ⊆ B

suppose if we have C = {0,1,2,3} and B = {0,1,4,6}

C ⊆ B Here C is not subset of B

Empty set is subset of every set and Every set is subset of itself

Ex: if A = { 1,2,3 } find all possible subset of set A

ɸ, {1}, {2}, {3}, {1,2},{1,3}, {2,3}, {1,2,3} Now Here total no. of Subset is 8

Here we can say if any set have nth element then the total no. of subset will be 2n
and

Subset are two types

Subsets are

- classified as

 Proper Subset
 Improper Subsets

Proper Subset Symbol

Let A and B are two sets, Set A is said Proper subset of set B if every element of set
A is also element of set B but But A = B

A proper subset is denoted by ⊂ and is read as ‘is a proper subset of’. Using this
symbol, we can express a proper subset for set A and set B as;

A⊂B

Set A is considered to be a proper subset of Set B if Set B contains at least one

element that is not present in Set A.

Example: If set A has elements as {12, 24} and set B has elements as {12, 24, 36},
then set A is the proper subset of B because 36 is not present in the set A.

A⊂B

But A = B
Ex: : if A = { 1,2,3 }

So the subset is ɸ, {1}, {2}, {3}, {1,2},{1,3}, {2,3}, {1,2,3}

proper subset = ɸ, {1}, {2}, {3}, {1,2},{1,3}, {2,3} and Improper set is {1,2,3}

Improper Subset : Every set is a subset of itself, called an Improper subset.

: if A { 1,2,3 }

Total subset ɸ, {1}, {2}, {3}, {1,2},{1,3}, {2,3}, {1,2,3}

Proper subset ɸ, {1}, {2}, {3}, {1,2},{1,3}, {2,3}, remaining subset

Improper subset {1,2,3} because again we’re getting

Ex : A = ɸ here we don’t find any subset so ɸ is Improper Subset

If a set has n element then total proper subset = 2n - 1

All Subsets of a Set

The subsets of any set consists of all possible sets including its elements and the
null set. Let us understand with the help of an example.

Example: Find all the subsets of set A = {1,2,3,4}

Solution: Given, A = {1,2,3,4}

Subsets =

{}

{1}, {2}, {3}, {4},

{1,2}, {1,3}, {1,4}, {2,3},{2,4}, {3,4},

{1,2,3}, {2,3,4}, {1,3,4}, {1,2,4}

{1,2,3,4},

Example: Find all the subsets of set X = {1,5,7,9}

Solution: Given, X = ɸ, {1} {5}, {7}, {9},

{1,5}, {1,7}, {1,9}, {5,7}, {5,9}, {7,9}

{1,5,7} {1,5,9} {1,7,9}, {5,7,9}

{1,5,7,9}

if A = {2,4,6}

Number of subsets: {2}, {4}, {6}, {2,4}, {4,6}, {2,6}, {2,4,6} and Φ or {}.

Proper Subsets: {}, {2}, {4}, {6}, {2,4}, {4,6}, {2,6}

Improper Subset: {2,4,6}

itertools.combinations():

The function itertools.combinations() is intended to provide every possible

combination of the items in a given iterable

The itertools.combinations() function in Python takes an iterable as its first

argument and an integer r as its second argument. The iterable can be any
object like a list, tuple, string, or set.

Here is a simple program to showcase the different iterators that can be used.

the combinations() function basically returns an iterator. We can loop through

the iterator to get our results. But, we can also convert it into a list and get all
the combinations generated through the iterator in a list. Let us look at the
example below to convert it into a list.
Example
import itertools

s = {1, 2, 3}
k=2

# Find all subsets of size k

subsets = list(map(set,itertools.combinations(s, k)))

# Print the subsets
print(subsets)
Output:
[{1, 2}, {1, 3}, {2, 3}]

Example :
x = ['A', 'B', 'C', 'D']
result = list(map(set, itertools.combinations(x, 2)))
print(list(result))

Output :
[{'A', 'B'}, {'C', 'A'}, {'D', 'A'}, {'C', 'B'}, {'D', 'B'}, {'D', 'C'}]

Example :
import itertools

def findsubsets(s, n):

return list(itertools.combinations(s, n))

s = {1, 2, 3}
n=2

print(findsubsets(s, n))
Output:
[(1, 2), (1, 3), (2, 3)]

Example

from itertools import combinations

def getsubset(s):
subset=[]
for i in range(len(s)+1):
for m in combinations(s,i):
subset.append(m)
return subset

set1=[1,2,3]
subs=getsubset(set1)
print(list(map(set,subs)))

Output:
[set(), {1}, {2}, {3}, {1, 2}, {1, 3}, {2, 3}, {1, 2, 3}]

Here nested loop are required :

print("\n")
for i in range(4):
print("\n")
for j in range(i+1):
print(j,end=" ")

Output:

0
01
012
0123

Improper Subset
A=B

Improper Proper
⊆ ⊂
A=B

Fuzzy Set

Union

Fuzzy Set

Fuzzy Logic is derived from fuzzy set theory

Many degree of membership (between 0 to 1) are allowed. Thus a membership

function is associated with a fuzzy sets à such that the function maps every
element of universe of discourse X to the interval [0,1].

The mapping is written as: (x): X → [0,1].

Fuzzy Logic is capable of handing inherently imprecise (vague or inexact or rough

or inaccurate) concepts

Example
Fuzzy Sets (Continue)

• Let X = {g, g, g, g, g} be the reference set of students.

• Let A be the fuzzy set of "smart" students, where "smart" is fuzzy term.

A = {(g1,0.4) (g 2,0.5) (g3,1)(g4,0.9)(g5,0.8)}

Here à indicates that the smartness of g, is 0.4 and so on

Property of Fuzzy Set

Fuzzy sets are defined as sets that contain elements having varying degrees of
membership values. Given A and B are two fuzzy sets, here are the main properties
of those fuzzy sets

A = {1, 2, 3},

B = {2, 3, 4},

C = {5, 6}

Commutativity :-

(A ∪ B) = (B ∪ A) A ∪ B = {1, 2, 3, 4} → LHS
(A ∩ B) = (B ∩ A) B ∪ A = {1, 2, 3, 4} → RHS
A ∩ B = {2, 3} → LHS
B ∩ A = {2, 3} → RHS
Associativity :- A ∪ B = {1, 2, 3, 4}
(A ∪ B) ∪ C={1, 2, 3, 4, 5, 6}
 (A ∪ B) ∪ C = A ∪ (B ∪ C) B ∪ C = {2, 3, 4, 5, 6}
 (A ∩ B) ∩ C = A ∩ (B ∩ C) A ∪ (B ∪ C) = {1, 2, 3, 4, 5, 6} → RHS

A ∩ B = {2, 3}
(A ∩ B) ∩ C = ϕ → LHS
B∩C=ϕ
A ∩ (B ∩ C) = ϕ → RHS

Distributivity :- A ∪ (B ∩ C) = (A ∪ B) ∩ (A ∪ C)
B∩C=ϕ
 A ∪ (B ∩ C) = (A ∪ B) ∩ (A ∪ C) A ∪ (B ∩ C) = {1, 2, 3}→LHS
 A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C) A ∪ B = {1, 2, 3, 4}
A = {1, 2, 3}, A ∪ C = {1, 2, 3, 5, 6}
(A ∪ B) ∩ (A ∪ C) = {1, 2, 3} → RHS
B = {2, 3, 4},
A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C)
C = {5, 6} B ∪ C = {2, 3, 4, 5, 6}
A ∩ (B ∪ C) = {2, 3} → LHS
 A ∩ B = {2, 3}
A∩C=ϕ
(A ∩ B) ∪ (A ∩ C) = {2, 3} → RHS

Idempotent :- Idempotency is defined as,

A∪A=A
A∩A=A A∪A=A

A∩A=A

For the given data,

A ∪ A = {1, 2, 3} = A

Identity :- Mathematically, we can define this

A ∪ Φ = A => A ∪ X = X property as,
A ∩ Φ = Φ => A ∩ X = A
A∪X=X
A∩X=A
A∪ϕ=A
A∩ϕ=ϕ
For the given data,

A ∪ X = {1, 2, 3, 4, 5, 6} = X
A ∩ X = {1, 2, 3} = A
A ∪ ϕ = {1, 2, 3} = A
A∩ϕ={}=ϕ
Transitivity :-

If A ⊆ B, B ⊆ C, then A ⊆ C

Involution :-

(Ac)c = A

µ = {1,2,3,4,5}

A={2,3}

Ac = µ - A = {1,4,5}
L.H.S. (Ac)c => {1,2,3,4,5} – {1,4,5} = {2,3}

Example

µ = {a,b,c,d}

A={a,b}

Ac = µ - A = {a,b,c,d} - {a,b} = {c,d}

L.H.S. = (Ac)c = {a,b,c,d} – {c,d} = a,b

(Ac)c = A

Sets are used to store multiple items in a single variable.

Set is one of 4 built-in data types in Python used to store collections of data, the
other 3 are List, Tuple, and Dictionary, all with different qualities and usage.

A set is a collection which is unordered, unchangeable*, and unindexed.

* Note: Set items are un changeable, but you can remove items and add new
items.

Union

set1 = {"a", "b" , "c",5.2}

set2 = {1, 2, 3}
t=set1.union(set2)
print(set1.union(set2))
print(t)

Output :
{1, 2, 'a', 3, 5.2, 'b', 'c'}

Example
set1 = {"a", "b" , "c",5.2}
0set2 = {1, 2, 3}
set1.update(set2)
print(set1)

Output:
{1, 2, 3, 'b', 5.2, 'c', 'a'}
typecast into set
x=["ab",4,9,12,3.5,12,9.2]
m=set(x)
print(m)
output :
{3.5, 4, 9, 9.2, 12, 'ab'}

Subset
a={1,2,3,}
b={1,2,3,4}
print(a.issubset(b))

Output:
True

Intersection
a={3,6,0,7.6,"a",1,False,15}
b={1,3,6,9,12,15,0}
z=a.intersection(b)
print(z)

output :
{0, 1, 3, 6, 15}

Intersection using loop

l1 = ['one', 'two', 'three']
l2 = ['one', 'six', 'three']
l3 = ['one', 'four', 'five']
l4 = ['one', 'three','five']

check_list = list(set(l) for l in (l2, l3, l4))

print(check_list)
print("\n")
result = set(l1)
for s in check_list:
result = result.intersection(s)
print(result)

output :
[{'three', 'one', 'six'}, {'four', 'one', 'five'}, {'three', 'five', 'one'}]

{'one'}
Fuzzy to Crisp Conversion

We have various method of defuzzification

1. Max – membership principal

2. Centroid method
3. Weighted average method
4. Mean – max membership
5. Center of sums
6. Center of Largest area
7. First of maxima, last of maxima

Max – membership principal

Also known as the height method, this scheme is limited to peaked
output functions. This method is given by the algebraic expression where
z∗ is the defuzzified value. ∫

Here membership value µ given in

highest peak on point which is the
highest position on 1 and highest
value on z axis is z* so z* is the crisp
value or defuzzified value by using
maximum membership principal

Centroid method

This method is also known as center of mass or center of area or center of

gravity which needs to find out defuzzyfied value.

Let take an example by taking the union of two fuzzy set so in this diagram two
axis y is µ membership value and x axis and defuzzified value received by
centeroid two fuzzy set the centroid value in between two fuzzy set
 Here centroid value is Z* which Gives the z point which is located
defuzzified value and here A~ is fuzzy at the center of gravity
set value
Let take another example

Here Set 3 is the union of two fuzzy set set1 and set 2

Syllabus

Unit - 1

This works on LLM (Large Language Model)

 LLM stands for Large Language Model, which is a type of Artificial
intelligence (AI) that can understand and generate human
language. LLMs are a subset of deep learning, a type of machine
learning that uses algorithms to recognize complex patterns in large data
sets

Some Basic Components of ANN Artificial Neural Network

• Neurons (Nodes)

• Layers

• Connections

• Activation Function

Now check the degree of fastness

Data is stored and it will calculating the data now it is creating Layets when its
processing the data

Layers Input data

Data Stores hidden layer data is breaking and associated

Layers are multiple Layers

Connection

The above picture is a Neuron Network

What Are Activation Functions?

Activation functions are an integral building block of neural networks that enable them
to learn complex patterns in data. They transform the input signal of a node in a neural
network into an output signal that is then passed on to the next layer.
An activation function in a neural network determines which neurons are
activated as information moves through the network's layers. It's a key
component of neural networks that allows them to learn complex patterns in
data.

Neural Network
A neural network is a machine learning technique that uses a network of
interconnected nodes to process data and learn from mistakes,
For example Internet : Sharing the Data
If we talk about human brain there are high number of neurons
neurone,[ HYPERLINK "https://en.wikipedia.org/wiki/Neuron"1 HYPERLINK "https://en.wikipedia.org/wiki/Neuron"] or nerve cell which are
connected
There are gazillions of Neurons in Human Brain which are connected with
network and that connectivity provides the flow information and process it
and parallel process
In Human Brain can learning algorithm from experience and train the data in
itself and the capacity of training and learning make the human brain a very
important organ
And we do in artificial machine we want to implement brain like human brain
which we can say neuron network and that we can say Artificial Neural
Network which works the learn the algorithm and train the data
For Example, First day you go to the office then you recognize the faces of
people and staff on there and some have less hair shaved or trim detect the
faces next again you go to office then you’ll detect the face and recognize the
faces so many people may have shaved or trimmed But you will recognize
them.
Human brain learns and trains But the same thing if we do in Normal
Computer then it’ll have to do all the scanning if there are 100 people. You’ll
have to scan their photos and give them to the computer. If you check again
the next day
 So the Normal computer can work only the base of algorithm which gives the
output like recognize the face expression of any actor
If we implement this with the help of Neural Network, then it recognizes 90%
so here the accuracy is increasing this is the power of Learning algorithm

Here this biological neuron we are implementing in Artificial Neurons in the

Machine Learning where we train the data
1 billion = 100 Crore or 1 crore = 100 lakh
10 crore = 1000 Lakh
So 1 billion = 100 crore = 10000 Lakh
1 trillion = 1000 billion
1 trillion
Llama 3
Well, first of all, Meta’s Llama 3 has a 15 trillion–token dataset (enabling
more efficient language encoding and better performance), which is 7x times
larger than previous models
In ANN the LLM (large language model) used with the help of token and
parameter
Token
Lets
Types of Artificial Neural Network
• Single Layer Feed Forwarding Network
• Multi Layer Feed Forward Network
• Multi Layer Perceptron Network
• Feedback ANN

• Single Layer Feed Forwarding Network1

There are only two layer Input and Output
• Multi Layer Feed Forward Network
There is Hidden Layer in Multi Layer Feed Forward Network
Means here we can solve complex programming and their al

Here nodes are connected with some weight and other layer nodes are
connected here with the help of activation function we can apply
corresponding input get the output

• Multi Layer Perceptron Network

 In this three or more layers classified data are used to classify
non-linearly separable data
This is fully connected node means each node connected to next
layer node and usage non-linear activation function

Let’s assume we have input x1, x2, x3 all will be connected and
rest of the layer will be connection completely after that we get
Output

4 Feedback ANN
Feedback is provided to adjust parameters
Here parameter return back towards first layer if any error occurs then
parameter can be updated and because to minimize
Non Linear Separable Data

Example 2
7, 7=> 1 (Pass),

2 hrs. sleep, 8 hrs. study => 0 (Fail)

Y =x1.w1 + x2.w2 + b

I want output come between 0 & 1 in y then use sigmoid function

Here b (bias) value will update as much as update then value will be
fluctuating it.

So at initial level we have train the data giving the value of w1 and w2 and
updated it so that loss should be minimal
 this is sigmoid function
w1=0.5
w2=-0.3
b1=0.1
Now check formula: Y =x1.w1 + x2.w2 + b1 let’s take new data for the training
the data

= 2*0.5 + 8*-0.3 + 0.1

1
-2.4
0.1
-1.3
Here -1.3 value comes after using activation function now put this in sigmoid
function and value
1/1+(e^-(-1.3)) = 1/4.67 => 0.21 means it’s the probability near fail
Lets say we take x1= 6 and x2 = 7 in activation function
Y =x1.w1 + x2.w2 + b1
= 6*0.5 + 7* -0.3 + 0.1 = 1
 x1 x2 Total
6 7 0.1
0.5 -0.3
3 -2.1 0.1 1
0.9 0.1 1

1/1+(e^-(1)) = 1/1.37 => 0.72 which means its probability near

Pass
Recurrent Neural Network (RNN)
It is used in Feed Forward Neural Network stage and they
cannot handle sequential data because they use only current
input while we need previous state and memory element is
absent

FFN RNN
Forward neural network which is going forward and it does not
have any memory
It takes current input but here we require previous state
which they do not consider it
To overcome this problem, we need RNN which have
memory and previous data and current input
It also use loop method means when output comes again it goes

input state
It also consider previous state but suppose it has not previous
state then it use0s at starting Initial Hidden State
Which is H0 and previous hidden state and X1 current input
both process it uses with first Time Step then output will be
Current Hidden State

ht= f(ht-1, Xt)

ht = current state
ht-1 = Previous state
Activation Function
ht= tanh(Whh .ht-1 + Wxh . Xt )
Whh = Weight at Previous Hidden State
Wxh = Weight at Current Input State
Yt = Why . ht
Why = Weight at the output state
ht = time step h1 and h2,h3….hn
A recurrent neural network (RNN) is a kind of artificial neural
network mainly used in speech recognition and natural
language processing (NLP). RNN is used in deep learning and
in the development of models that imitate the activity of neurons
in the human brain.
Recurrent Networks are designed to recognize patterns in data
sequences, such as text, genomes, handwriting, the spoken
word, and numerical time series data emanating from sensors,
stock markets, and government agencies.
Genmo AI is a tool that uses artificial intelligence (AI) to create
videos and images:

What is Deep Learning

Deep Learning is a form of Machine Learning. It is known as

'Deep' Learning because it contains many layers of neurons. A
neuron within a Deep Learning network is similar to a neuron of
the human brain - another name for Deep Learning is
'Artificial Neural Networks'.
Various Learning Techniques of ANN
 • Supervised Learning

• Unsupervised Learning

• Reinforcement Learning f

Supervised Learning

Our main is that how to make intelligent the machine in supervise learning e
have training data means supervise means supervisor (teacher) and we have
input data and output data which we create a model and that model we put
new input and check valid output Example : Exit poll in Election

Now here learning algorithm works like Naïve Bayes input your
algorithm is learning and classifying means it will check the
probability of wining or loosing so this is supervised
learning
As the name suggests, this type of learning is done under the
supervision of a teacher. This learning process is dependent.
During the training of ANN under supervised learning, the input
vector is presented to the network, which will give an output
vector. This output vector is compared with the desired output
vector. An error signal is generated, if there is a difference
between the actual output and the desired output vector.
On the basis of this error signal, the weights are adjusted
until the actual output is matched with the desired output.
Used Learning Algorithm Naïve Bayes Algorithm
Which apply input and output given and again given New
Input actual counting in exit poll and prepare in Model
And find out result probability or output
Naïve Bayes theorem usage in Supervised Learning
Where we need to predict for any result for example Email is
spam or Not spam here we need to classify which requires
Bays theorem

Bag 1 Total Red Ball Probability in 1st Bag = 2/5

Bag 2 Total Red Ball Probability in 2nd Bag = 4/7

Now suppose we want to take a One Ball which should be

Red from Both Bags. Then total probability will be

1/2 x 2/5 + 1/2 x 4/7

Now comes Bays Theorem I have taken a Ball which is

Red

What is the probability if the Red ball is taken out from Bag1
-> (B1/R) so this is called cause property means Reverse
method

Red ball is taken from Bag1 means Ball is Given Red from
Bag1 -> 1/2 X 2/5-> B1/R
So formula is Bays Theorem P(Y/X) = P(X/Y) * P(Y) / P(X)

Here X is given that Red Ball and find out of Red is taken
from Bag1

= R/B1 * B1

P(Y/X) = P(X/Y) * P(Y) = P(R/B1) * P(B1)

P(X) P(R)

= 2/5 * 1/2

1/2 x 2/5 + 1/2 x 4/7

= 1/5

1/5+2/7

Suppose this is Yes, Red ball taken from B1

P(Y/X1,X2…….Xn) = P(X1/Y)*P(X2/Y)*P(X3/Y)……P(Xn/Y)*P(Y)

P(X1) *P(X2)*P(X3)……P(Xn)

Again Suppose this is No, Red ball is not from B1

P(N/X)= P(X/N)*P(N)

P(X)

P(N/X1,X2…….Xn) = P(X1/N)*P(X2/N)*P(X3/N)……P(Xn/N)*P(N)

P(X1) *P(X2)*P(X3)……P(Xn)
Suppose Fever was Yes and We have Two Factor X1, X2

X1=Covid , X2=Flu

And Here person(Flu, Covid) -> X1,X2

P(X1/Y)*P(X2/Y)
P(Y/X1,X2)= P(X1/Y)*P(X2/Y) * P(Y) / P(X1) *P(X2)……..n

=4/7*3/7* 7/10

=12/70

Unsupervised Learning

As the name suggests, this type of learning is done without

the supervision of a teacher. This learning process is
independent. Here data is learning and we have only input
which make some cluster and refine that based on the any
category maximum machine are Unsupervised where we
have to input to supervised technique
During the training of ANN under unsupervised learning, the
input vectors of similar type are combined to form clusters.
When a new input pattern is applied, then the neural
network gives an output response indicating the class to
which the input pattern belongs.
Reinforcement learning
It works in Reward and Policy
It is done by Action and changed the state get the
reward
Suppose given Rs. 50 and change the state and again
retrieve the reward then create any policy
Pk Movies takes currency behalf of Gandhi Note Ji and realize
the penalty and learn the data
It becomes in Game that we can create in the for next Level

In Supervised Data we give input and give Output then we find

our desired output
Continuous Data like we give similar input for weather forecasting
morning (Humidity) afternoon (Hot) and Evening (Cloudy) then
machine will learn and give the prediction which is Label Data
here we give the Label Data Means Input Output Pair this is
Supervised Learning has two type of Learning Algorithm

1. Regression Algorithm

What is Temperature going to be tomorrow

Temperature means Values is continuous change

Regression Algorithm are used if there is relationship between

Input Variable and Output Variable

Example Age, Salary, Prices etc. means continuous change

Figure: Regression Analysis

Regression Analysis Algorithm :

Linear Regression, Logistic Regression, Support Regression,

Decision Tree Regression, Random Forest Regression

Linear Regression

Y=mX+b

Y represent the dependent Variable //to check score

X represent the independent Variable//Study Hrs.

m is the slope of the line (how much Y changes for a unit change
in X)

b is the intercept (the value of Y when X is 0)

2. Classification Algorithm

Here we need to classify the data Either Hot or Cold

Will it be cold or Hot Tomorrow?

Means Yes/No, True/False, Male/Female

Example : Linear Regression

Project : Predicting Pizza Prizes

Step 1 : Data Collection

Diamete Price Mea Mea Deviatio Deviatio Product Sum of Square

r (X) in (Y) in n (X) n (Y) n (X) n (Y) of Product of
Inches Dolle Deviatio of Deviatio
r n Deviatio n for X
n
8' 10 10 13 -2 -3 6 12 4
10' 13 0 0 0 0
12' 16 2 3 6 4

X is Independent and Y is dependent

Mean (x) = 8+10+12=30/3=10

Mean (Y) = 10+13+16=39/3=13

Sum of Product of Deviation =6+0+6=12

Now calculate m i.e, slope

m=Sum of product deviation/sum of square of Deviation for x

=12/8=1.5

Means if I am changing x value 1 then Y will be change 1.5

Now suppose pizza X size is 0 then what will be charged of Pizza

 Y = mX+b

Now Calculate b = Mean of Y – (m*Mean of X)

=13 – (1.5 * 10)

= 13 – 15 = -2

Suppose if we have 20’ size pizza then what will be Price

Y=mX+b

=1.5*20+(-2)

=30-2=28$

Regression analysis is a key part of predictive modeling

and is used in many different applications of machine
learning. For example, regression analysis can be used to:

• Predict house prices

• Forecast stock or share prices

• Map salary changes.

Different type of Classification Algorithm

• Decision Trees

• Random Forest

• Logistic Regression

• Support Vector Machine

Decision Trees

Random Forest

Here are multiple decision tree exist

Output is based on majority voting that email is either Spam or
Not Spam

R F is an ensemble learning method

The Algorithm which implement the classification on a data set is

known as a classifier

Random Decision forest correct for decision tree’s habit of

overfitting to their training set.

These are two type Classification

• Binary Classifier

If the classification problem has only two possible

outcomes then it is called as Binary Classifier

Example : Yes/No, Male/Female, Spam/Not Spam,

Cat/Dog, True/False

• Multiclass Classifier :

if a classifier problem has more than two outcomes

then it called as Multiclass Classifier

Example : Classification of types of crops,

Classification of types of Music

•
Non linear Regression

Support Vector Machine

Labeled data given in supervised algorithm for training data
If New data belongs to Square shape or Circle Shape during the
testing for Prediction
If the New data is circle, then output belongs to circle class
Linear SVM: Linear SVM is used for linearly separable data,
which means if a dataset can be classified into two
classes by using a single straight line, then such data is
termed as linearly separable data, and classifier is used called
as Linear SVM classifier.
Non-linear SVM: Non-Linear SVM is used for non-linearly
separated data, which means if a dataset cannot be
classified by using a straight line, then such data is termed
as non-linear data and classifier used is called as Non-linear
SVM classifier.
Unsupervised Learning
As the name suggests, this type of learning is done without the
supervision of a teacher. This
learning process is independent.
Here we give only input in model here checks hidden information
And it checks in pattern and recognise it then it creates cluster
And Group
Age Salary
22 10000
23 12000
24 14500
29 20000
31 22000
 35 25000
42 34000
48 40000
54 48000
59 65000

we don’t give label data I/O Here pattern are checked with
cluster suppose we give different shape in input like triangle
which needs to recognize the pattern in unsupervised from
other shapes now when we give new
shape than it does not match with other shape then this
rectangle shape puts into that group
 In the real world, we do not always input data with the
corresponding output so to solve such cases we need
unsupervised learning
in this learning use K-mean clustering

• is your data labeled or unlabelled? Supervised learning

requires labeled datasets. You’ll need to assess whether your
organization has the time, resources, and expertise to validate
and label data.
• What are your goals? It’s important to consider the type of
problem you’re trying to solve and whether you are trying to
create a prediction model or looking to discover new
insights or hidden patterns in data.
There are two type of Unsupervised Learning
• Clustering – making an group after separation
in different objects
• Association
 An Association rule is set an unsupervised learning
method used for finding the relationship between
variables in the large database
It determines the set of items that occur together
the dataset

Example :
Market Basket Analysis :

Reinforcement Learning
Reinforcement Learning is a feedback-based machine learning
Technique in which an agent learns to behave in an environment
by performing an action and seeing the result. For a good action,
the agent gets positive feedback, and for a bad action, the agent
gets negative feedback or a penalty.
There is no labeled data. So the agent is bound to learn by
its experience
Reinforcement Learning solves a specific type of problem where
decision-making is sequential and the goal is long-term, such as
game-play robotics.
Advantages of supervised
With the help of supervised learning the model can predict on the
basis of prior experiance
In Supervised Learning, we can Have an exact idea about the
class of the object
Supervised Learning model helps us to solve various real world
problem such as fraud detection, spam filtering
Disadvantages
Supervised Learning models are not suitable for complex task
If the test data is different from the training data then it can not
predict the correct output
Training required lots of computation times
We need enough knowledge about class of objects means images
of Fruits and their label
Perceptron Rule

Understanding the Perceptron

A perceptron is a type of artificial neural network invented in
1958 by Frank Rosenblatt. It is the simplest form of a neural
network, used for binary classification tasks.
Perceptron is a linear binary classifier used for supervised
learning.
Here are the key components and concepts related to a
perceptron:
Input Nodes: These are the features of the input data. Each
input node is assigned a weight.
Weights: Each input is multiplied by a weight which can be
adjusted during the learning process to minimize error.
Summation: The weighted inputs are summed together.
Activation Function: The sum of the weighted inputs is passed
through an activation function. In the case of a perceptron, a
step function is often used. The output of this function is the
prediction of the perceptron.
Bias: An additional parameter that allows the activation function
to be shifted to the left or right, improving the model’s fit.

Perceptron Rule is Supervised Learning algorithm because

we have supervision of output mapping
Perceptron are binary classifier which converts the data in
two classes
It is mapping of I/O in x, y

X Y
 Y = X1W1+X2W2…..XnWn +b
Step function : Y > 0 -> 1
Y < 0 -> -1
So this is also activation function
error e(t) = Y (desired o/p) – Y (Actual o/p)
Set the initial value is w1, w2………. Wn
In Supervised Learning, we have input and also output which
is labeled data or training data
On the basis of training data, we create a Model and again
give new input and check the output
q

On the basis of Label Data/Training (Input & Output) data, an

Error signal generated if difference between Actual Output and
Desired Output
On the basis of Error Signal weights are adjusted until both
Actual output is matched with Desired Output
Linear Regression

Y=mX+b

Y represent the dependent Variable //to check score

X represent the independent Variable//Study Hrs.

Activation Functions in Neural Networks

Sigmoid, Hyperbolic Tangent Function (Tanh),

Softmax, softsign function

Need of bias

Now, Suppose if b was absent, then the graph will be formed like this:

Due to absence of bias, model will train over point passing through origin
only, which is not in accordance with real-world scenario

w1=0.5
w2=-0.3
b1=0.1
Why we use Activation functions with Neural Networks?

It is used to determine the output of neural network like yes or no. It

maps the resulting values in between 0 to 1 or -1 to 1 etc. (depending
upon the function).

The Activation Functions can be basically divided into 2 types-

• Linear Activation Function

• Non-linear Activation Functions

Fig: Linear Activation Function

 Fig : Non Linear Activation Function

nucleus in anatomy is a brain structure (plural = nuclei). It is a

compact cluster of neurons.

Yes a perceptron (one fully connected unit) can be used for regression.
It will just be a linear regression. If you use no activation function you
get a regression and if you put a sigmoid activation you get a
classifier

Perceptron is a binary classifier

And Perceptron divides the data into two region means it is Line
And Perceptron will classify the Linear data means which is
separable Data through a Line
Y = f(z)
Y =X1.w1 + X2.w2 + b

w1=A, w2=B, b=C,

X1=x, X2=y
Here AX + BY + C is line
Ax + BY + C > 0 True then

f(z) = w1.x1+w2.x2+w3.x3 + b
ax+by+cz+b > 0

Linear Regression

Y=mX+b
 • Linear summation of inputs: In the above diagram, it has

two inputs x1, x2 with weights w1, w2, and bias b. And the

linear sum z = w1 x1 + w2 x2 + … + wn xn + b

This Linear Activation Function

Regression Analysis Algorithm :

Linear Regression, Logistic Regression, Support Regression,

Decision Tree Regression, Random Forest Regressy the data on

Logistic Regression : AX1+BX2+C

Suppose if we have extra line in graph. Then we can use

AX1+BX2+CX3+D

In perceptron we use Line that we want to search line which

could classify that is A, B, C means we pick random value for a
line to get the maximum probability

which we take a random value

0 => placed student

X => Not placed student
2D
Ax1+Bx2+C=0
3D
Ax1+BX2+Cx3+D=0
Perceptron Rule shows :

In Perceptron if X input and Weight vector are declared then

Perceptron is a type of linear classifier, while logistic regression is
a classification algorithm that can predict probabilities
P =X1W1+X2W2…..XnWn +b
P = ∑Wi * Xi
P=1, if P > 0
P=-1 if P < 0
P=0, if P=0

Perceptron Rule
in Perceptron we use Loop
Our goal is to find the line in the Graph means We have to find
out ABC Value to classify the data
• We start A=1, B=1, C=0 from Random Value
• Run Loop and
• 3
• Select a point or student
 • Check the point in the positive region or classified data
• Again check if it is a misclassification
• Change the ABC value and to make classification data
• Again repeat step no. 3,4 and 5 then go to step no. 6

Loop can be run in 1000 times or the second option till while
convergence check how many points are misclassified until it
becomes zero the loop will be executed after that it be
terminated
And the convergence done successfully loop will be terminated
Here we have convergence and another option epoc means no. of

To classify the x data and 0 data use loop in perceptron trick

 Until the data classify may be loop will be 1000 time either
10000 loop
In every loop we select random student that if the green
student is in correct position according to line then no changes in
A,B,C value
Now How to identify Positive Region and Negative Region

Now how to identify positive reason and region for this we

take tool https://www.desmos.com/calculator
2x+3y+5 =0

Again given new line coordinate value

2X+3y+5>0 then it gives positive reason
 Now u can check point whether it is positive reason or not
we know line equation : Ax+By+C=0 suppose these two
points coordinates x1 and y1 then we put x1and y1 into equation
then eq: Ax1+By1+C if it is > 0 then point is in Positive
reason else negative reason and it is = 0 then point is over the
line
 -2x+3y-5>0 then it is positive region and -2x+3y-5<0 then
negative region
If the value is Increased -> Down line
If the Value is Decreased -> Up line
Now here how the line is moving when C value is increasing
then line is going down and decreasing then it is up side
 X Value
2x+3y+5=0
4x+3y+5=0
x+3y+5=0

here x and y value is changed suppose your line is 2x+3y+5=0

and blue line coordinate value is (4,5) then add 1 -> 4,5,1
 and subtract it
Y Value
2x+3y+5=0
2x+6y+5=0
2x+y+5=0

New right side blue point is wrong position

to correct this take coordinate value i.e, 4,5 and add 1 which is
4,5,1 and subtract from 2 coefficient 5 which -2 -2 4 i.e,
-2x-2y+4=0

Again check green point which is negative region so now

apply 1,3,1 and add 2,3,5 =3x+6y+6=0

right side coordinate value 5,2 and left side -3,-2

Dvsdvs
-3x+y+4=0

Auto Associative Memory Network: it is obtained by its content

If any content is missed, then associative memory completes this
task
This is Single Layer architecture
Auto Associative Memory

This is a single layer neural network in which the input training

vector and the output target vectors are the same. The weights
are determined so that the network stores a set of patterns.

Associative memory is also known as content addressable

memory (CAM) or associative storage or associative array.
Training algorithm
Input Data is Vector Input “S” and Output Data Vector Output “T”
Will be same output S=T
Input will be partial information and Noisy Information Output
will be same

For training, this network is using the Hebb or Delta learning rule.
Step 1 − Initialize all the weights to zero as wij = 0, i=1 to n, j=1 to n
Step 2 − Perform steps 3-4 for each input vector.
Step 3 − Activate each input unit as follows −
xi=si(i=1ton)xi=si(i=1ton)
Step 4 − Activate each output unit as follows −
yj=sj(j=1ton)yj=sj(j=1ton)
wij(new)=wij(old)+xiyjwij(new)=wij(old)+xiyj
Hebb rule :
Calculate the New Weight value
Activate each input unit as follows −
xi=si(i=1 to n) xi=si(i=1 to n)
Step 4 − Activate each output unit as follows −
yj=sj(j=1 to n) yj=sj(j=1 to n)
Step 5 − Adjust the weights as follows −
wij(new)=wij(old)+xiyj
Outer Product Rule
Wn = [S]t * [T]
t=transpose matrices, T=Target Matrics
Check the auto-associative network for input vector [1 1 -1] from
the weight vector with no self connection. Test whether the net is
able to recognize with one missing entry.
Input vector x = [1 1 -1]
And here no self connection in Weight -> No W i Which is 0
Here St= target input and transpose of input
W = ∑ST x t = |1|
|1| [1 1 -1]
|-1|

[1x1 1x1 1x-1]

[1x1 1x1 1x-1]
[-1x1 -1x1 -1x-1]

[1 1 -1] [0 1 -1]
[1 1 -1] = [1 0 -1]
[-1 -1 1] [-1 -1 0]

In weight No Self connection i.e, 0 so matrix will diagonal

value zero

Now Test Input with one missing entry so X = [1 0 -1]

[0 1 -1]
Y = x.W = [1 0 -1] [1 0 -1]
[-1 -1 0]

[1x0+0x1+-1x-1 1x1+0x0+-1x-1 1x-1+0x-1+-1x0]

[0+0+1 1+0+1 -1+0+0]

[1 2 -1]

Hetro Associative Memory

It is capable of retrieving piece of data one category upon
presentation of another category.
Here one category given input then it gives similar type output
Here Output Ym and input Xn
SO input set S not equal to Output T (Target data set)

Gradient Descent
Gradient Descent is an optimization algorithm which means to find out
best result for the algorithm and it is a optimization technique if we
give differentiable function then gradient function will give the minimal
result
here i have taken data set suppose here cgpa lpa of four student
cgpa lpa

Real world data affects the real world factor suppose student not
place in desired lpa may be he was not suited or qualified
interviewed which is called stochastic error
and due to stochastic error data becomes sort of linear
Backpropagation
It is a algorithm to train neural network
This is a supervised learning algorithm used for training a
deep-learning model
This learning method is the most popular at the moment
because it makes possible the use of powerful calculation
with low computation time
Backpropagation Algo => Train NN
In backpropagation we check errors after the calculation
of the Linear Activation Function

Y = X1W1+X2W2…..XnWn +b

In Neural Network has two thing :

Weight and Bias so which we have to calculate
Before calculation we should know two algorithm
• Gradient Descent
• Forward Propagation
Suppose we found 18 LPA for first student, after calculation of the
Linear activation function
But It is given 3 LPA now we use the Loss function

Here X vector input data and Y Vector our data are not same.

Step 1 : Initialize W=1 and b=0 value

Step 2 : Select a point row or student
Step 3 : Choose a loss function and calculate mean squared error MSE
which is given below :
Where y is actual data and is prediction data

L = (3-18)2
= 225
Now we have to reduce error because Neural Network is
3 LPA here actual data output y i.e, 3 and ŷ i.e, 18 which
prediction of Neural Network
ŷ Here is O21
O21 = W211 . O11 + W221 . O12 + b21

Here ^Y i.e, O21 is dependent on 5 input

ŷ=
Similarly, b11 i.e, O11 is dependent on IQ, CGPA, W111, W121 and b11 on
it self.

Now can these are complex hierarchy so If I have to minimize loss then
we can change the value of bias and weight value but we cannot
change cgpa and iq value because that is our data
So to minimize and decrease the loss then we go to previous or back
step to change the value of weight and bias which is called
backpropagation
Backpropagation Error= Actual Output – Desired Output
Because we are approaching to solve this error after go to back step to
update the value of Weight and Bias using with Gradient descent
Step 4 : Here update the weight and bias value using with gradient
descent
W new = W old – Learning rate times partial derivative of loss
function with respect to weight
Similarly, it is applicable in b value
Gradient descent
Gradient descent is an optimization algorithm that's used when training
a machine learning model.

We have data and we’ll create best fit line

In a graph, a residual is the difference between the actual value of a
data point and the predicted value of a data point:
 Create best-fit line for linear
regression
Yi= actual output
n=4 student

Here loss function is dependent

on m and b
If we change the m and b value
then
Suppose we also know m value
m=78.35

Best fit line means which can minimize the error on y direction
The sum square of Difference of Actual y value and predicted y value
should be minimum
M is the slop in the line
Here we make a simple discussion and assumes m=78.35
n
L= ∑ (yi - ŷ)2
i

Here we need b value which is minimum for L value Here L and b

dependent on square L->b2 so we create a graph so it is parabolic graph
1
L

Here we increase b then it close

to L point
And we decrease b value then it is
far from minimum of L

Step 1 :In Gradient Descent select random value of b let assume b= -10
set the b value to reach minimum value at L
How do we know to check L minimum after decreasing or increasing of
b so we can find it from slope
 How to find out a particular slope of a function let assume we have
function

y=x2 + 2 and x=5 given and what is slope then we differentiate dy/dx=
2x
so dy/dx = 10 this function belongs to 10 on the slop
suppose when we find slop and if the slope is negative then we move
forward means increase the b value and if the slope is positive then
go backward and decrease the b value so if you want b new value
b new = b old – slope
this is gradient descent
initially we had b=-10 and suppose slope = - 50
then b new = -10 –(-50)
b new = 40 which you move forward
L

b
again we take b value suppose b=b10 and slop = 50
b new = b old – slope
= 10 – 50 = -40
L

b
L L

30 40 b b

b new = b old – slope b new = b old - ƞ slope

b new = -10-(-50) =40 b new = -10 – (0.01*-50)
Here it makes large jump so use = -10 + 0.5 = -9.5
now create the transform Again iterate = -9.5 –(0.01* -40)
equation with Learning Rate ƞ = -9.5 + 0.4 = -9.1

move backward
Linear Regression
Slope -> m= tanθ
Simple Linear Regression
This is the simplest form of linear regression, and it involves only one
independent variable and one dependent variable. The equation for
simple linear regression is:
y=mX+β
where:
 Y is the dependent variable
 X is the independent variable
 β is the intercept
 m is the slope

Suppose we have data x and y

X= cgpa and y= lpa

1. This is sort of linear 2. This is completely linear

3. 4. put best fit line

Step 1 : Initialize W=1 and b=0 value

Step 2 : Select a point row or student
Step 3 : Choose a loss function and calculate mean squared error
MSE which is given below :
L=(yi-ŷ)^2
Step 4 : Here update the weight and bias value using with
gradient descent

Types of Gradient Discent

 Batch Gradient Discent
 Stochastic Gradient Discent
 Mini Batch Gradient Discent

Suppose a student could not qualify enough lpa after having maximum
cgpa what is reason then why does it happen because real world data
affects from real factors which u can not understand by
mathematically what was reason for this lpa which you can not
qyantify so this called stochastic error
And due to stochastic error, which we can not determine and data
becomes sort of linear
 3. Suppose if we have sort of linear data then we put perfect line
and touch to the point which is best fit line

Now we’ll check m and b in next topic

(m, b)

closed Non closed

form form

OLS Gradient
Descent
Linear
SGD
Regression
Figure 1 Regressor
class

In Figure 1 Here need the correct value to balance m and b which

cover minimum error on this line
Closed form means a mathematical formula i.e,
ols b= y̅ - mx̅

Here x̅, y̅ is mean

value xi current
cgpa and yi is current row package
Figure 2
SGD : (Stochastic Gradient Descent) In
Figure 1 Here need the correct value
to balance m and b which cover minimum error on this line
Here in OLS ordinary least square we have also formula
In Figure 2 shows an error because line is touched on every cgpa
which shows distance from line
E= d1+d2+d3….+dn : in total
error here + and – will cancel
because some point in up side
and down side
E= d21+d22+d23….+d2n and if it is
done in square then both will be
in positive
E= 𝑑

2
where di is distance and y^ is the prediction 4
output and yi is the actual output so the this d
shows the difference with error for a particular
student and total error is which add all errors.
Total Error is for all student
3
Multiple Linear Regression
Here multiple input are required for this problem.
let x1,x2,x3 are input and independent data and y is output dependent
so multiple linear regression is just extension of simple linear
regression and in reverse order simple linear regression is just
specialization of multiple linear regression
let takes example for 100 students and we have cgpa, iq, gender, lpa
Cgpa Iq Gender Lpa
x1 X2 X3 Y
1
2
.
.

100

Here data is 4D 3 input and 1 output

while y= mx+b is 2D or we can say y = β0+β1X here β0 is b and β1 is m
Hyperplanes are represented by equations and can be used to classify
data points based on their position
now here we find out the hyperplane of 4D data in the equation

y=β0+β1X1+β2X2 +β3X3 Suppose here we want predicted value means

and now
This involves more than one independent variable and one dependent
variable. The equation for multiple linear regression is:
y=β0+β1X1+β2X2+………βnXn
\where:
 Y is the dependent variable
 X1, X2, …, Xn are the independent variables
 β0 is the intercept
 β1,β1 …, βn are the slopes

The goal of the algorithm is to find the best Fit Line equation that can
predict the values based on the independent variables.
X1 : cgpa X2 : iq Y

here data is 3 D
LPA

** ***
******
** * * * * *

cgpa
IQ

in 2 D y= mx+b
here in 3D formula y= mx1+nx2+b
y=β0+β1X1+β2X2 β0, β1, β2
here in 4D formula
y=β0+β1X1+β2X2+β3X2 β0, β1, β2, β3
if Data Set n dimensional
y=β0+β1X1+β2X2+β3X2……….+βnX2

n
y = β0+ ∑ βiXi
i=1

if n=1
y=β0 + β1x1
so linear regression we try to find out coefficient value if we have 1
input column then we find out 2 coefficient value
and if we have 2 input column then we find out 3 coefficient value
if we have n input columns then we find out n+1 coefficient value
earlier we have 3D data set
cgpa –x1, iq -> x2, y ->lpa and suppose if you have value of β0, β1, β2
y=β0+β1X1+β2X2 here if you want to calculate lpa
lpa = β0 + β1* cgpa + β2*iq
so if we can find out lpa for a new student then above formula we can
evaluate it.
here coefficient represents like weight in the above formula β2
mentioned what is the weightage of IQ to calculate the LPA
if β1 > β2 then we can imagine cgpa plays an important role as
compared to iq if lpa is calculated
β0 is offset means intercept value and if cgpa =0 and iq = 0 then it is
decided by intercept values to calculate the LPA
calculate 100 student predicted lpa ŷ
1 student Cgpa x1 IQ x2 Gender x3