INDUSTRIAL TRAINING REPORT

On

Course “Artificial Intelligence”

Undertaken at

Udemy
Submitted in partial fulfillment of the requirements for the award of degree of

Bachelor of technology
In

Electrical and Electronics Engineering

Submitted by

SIDDHANT GUPTA

Bhagwan Parshuram Institute of Technology

Affiliated To GGSIPU, Delhi
 DECLARATION

I, SIDDHANT GUPTA, Student of B.Tech (EEE) declare that the report titled “ARTIFICIAL
INTELLIGENCE” which is submitted by me to Department of ELECTRICAL AND ELECTRONICS
ENGINEERING, BHAGWAN PARSHURAM INSTITUTE OF TECHNOLOGY DELHI affiliated to
GURU GOBIND SINGH INDRAPRASTHA UNIVERSITY.

I also declare that this report is my original work and does not breach any existing copy- right.
This particular report has not been previously submitted to any other
University/College/Organization for academic qualification/ certificate/ diploma or degree.

Date:

SIDDHANT GUPTA

1
 ACKNOWLEDGEMENT

The successful completion of this course mark the beginning of an ever - going learning
experience of converting ideas and concepts into real life. This course was a quite a learning
experience for me at each and every step. At the same time it has given me confidence to work
and learn more in the field of artificial intelligence. I feel the experience gained during the course
will lead me to gain the bright prospect in the future.

I would like to give thanks to Udemy (online learning platform) and the course instructor there
who gave me all this valuable knowledge.

I express my sincere thanks to professor BS Pali (EEE Department BPIT) who introduced me to
Udemy , and for his active support and continuous guidance.

SIDDHANT GUPTA

40620804918

BTech(EEE)

2
CERTIFICATE

3
 CONTENTS

DECLARATION ................................................. 1
ACKNOWLEDGEMENT .................................... 2
CERTIFICATE .................................................. 3

Introduction

Artificial Intelligence ................................................... 6

Definition .............................................................6
History ................................................................ 7
Development of AI .............................................. 9
Why AI ?............................................................10
Use of AI .......................................................... 11

Introduction to Reinforcement learning ..................... 12

Bellman Equation ............................................. 13
Markov Decision Processes ..............................15
Plan & Policy .....................................................17
Living Penalty ................................................... 17

4
Deep Learning ........................................................... 18
The Neuron ........................................................ 18
Working of Neural Network ................................ 19
How Neural Network Learn ................................ 20
Activation Function ............................................. 20
Gradient Descent ............................................... 22
Stochastic Gradient Descent .............................. 23
Back Propagation ................................................24
Module in Course ....................................................... 25

ANNEX
Software & Language… .............................................27
Programming Languages… ................................ 27
Python .......................................................... 27
What is Anaconda…............................................ 29
Spyder… ......................................................29

5
 What is Artificial intelligence (AI)

The term AI was first coined the term in 1956

By John McCarthy.

Artificial intelligence (AI) refers to the

simulation of human intelligence in machines
that are programmed to think like humans and
mimic their actions. The term may also be
applied to any machine that exhibits traits
associated with a human mind such as learning
and problem-solving.

The ideal characteristic of artificial intelligence

is its ability to rationalize and take actions that
have the best chance of achieving a specific
goal.

Artificial intelligence is based on the principle

that human intelligence can be defined in a way
that a machine can easily mimic it and execute
tasks, from the most simple to those that are
even more complex. The goals of artificial
intelligence include learning, reasoning, and
perception. John McCarthy
AI is continuously evolving to benefit many
different industries. Machines are wired using a One of the greatest innovators in the field
cross-disciplinary approach based in was John McCarthy, widely recognized as
mathematics, computer science, linguistics, the father of Artificial Intelligence, who on
psychology, and more. the 24th of October, 2011, passed away at
the age of 84. He leaves behind him a great
legacy in the field of Computer Science and
Artificial Intelligence (AI)

6
 History
The idea of inanimate objects coming to life as intelligent beings has been around for a long
time. The ancient Greeks had myths about robots, and Chinese and Egyptian engineers built
automatons.

The beginnings of modern AI can be traced to classical philosophers' attempts to describe human
thinking as a symbolic system.

Getting Serious About AI Research

In 1951, an machine known as Ferranti Mark 1 successfully used an algorithm to master checkers.
Subsequently, Newell and Simon developed General Problem Solver algorithm to solve

mathematical problems.

In the 1960s, researchers emphasized developing algorithms to solve mathematical problems and
geometrical theorems.

AI Winters

However, despite this well-funded global effort over several decades, computer scientists found it
incredibly difficult to create intelligence in machines.

New Millennium, New Opportunities

In the late 1990s, American corporations once again became interested in AI.

7
 IN A NUTTSHELL

Well the term AI was coined in 1956. But there was not a significant growth in the field. So was AI at
that time a bad invention or useless.
The answer is the lack of technology that can facilitate the needs of AI.

For an AI to work it needs data and a lot of data. To processes lots of data we need enormous
processing power.

Storage Comparison

In 1956 we used to rent storage device that needs fork lifts and air plane for transportation size was
about two refrigerators and weighted about a ton and were only available for renting for about $25000
for 5MB of storage.

In 1980 we got storage devices that were much smaller than the previous, storage capacity of 10MB,
can be bought for about $3,495.

Nowadays we can about 256GB for about $150, and are small enough to set on our fingertips.

8
DEVELOPMENT OF AI

Early AI research in the 1950s explored topics like problem solving and symbolic methods. In
the 1960s, the US Department of Defense took interest in this type of work and began training
computers to mimic basic human reasoning. For example, the Defense Advanced Research
Projects Agency (DARPA) completed street mapping projects in the 1970s. And DARPA
produced intelligent personal assistants in 2003, long before Siri, Alexa or Cortana were
household names.

This early work paved the way for the automation and formal reasoning that we see in computers
today, including decision support systems and smart search systems that can be designed to
complement and augment human abilities.

1950s–1970s 1980s–2010s Present Day

Neural Networks Machine Learning Deep Learning

 1950s–1970s - Early work with neural networks stirs excitement for “thinking machines.”
 1980s–2010s - Machine learning becomes popular.
 Present Day - Deep learning breakthroughs drive AI boom

9
 Why is artificial intelligence?

 AI automates repetitive learning and discovery through data. But AI is different from
hardware-driven, robotic automation. Instead of automating manual tasks, AI performs frequent,
high-volume, computerized tasks reliably and without fatigue. For this type of automation,
human inquiry is still essential to set up the system and ask the right questions.

 AI adds intelligence to existing products. In most cases, AI will not be sold as an individual
application. Rather, products you already use will be improved with AI capabilities, much like
Siri was added as a feature to a new generation of Apple products. Automation, conversational
platforms, bots and smart machines can be combined with large amounts of data to improve
many technologies at home and in the workplace, from security intelligence to investment
analysis.

 AI adapts through progressive learning algorithms to let the data do the programming. AI
finds structure and regularities in data so that the algorithm acquires a skill: The algorithm
becomes a classifier or a predictor. So, just as the algorithm can teach itself how to play chess, it
can teach itself what product to recommend next online. And the models adapt when given new
data. Back propagation is an AI technique that allows the model to adjust, through training and
added data, when the first answer is not quite right.

 AI analyzes more and deeper data using neural networks that have many hidden layers.
Building a fraud detection system with five hidden layers was almost impossible a few years ago.
All that has changed with incredible computer power and big data. You need lots of data to train
deep learning models because they learn directly from the data. The more data you can feed
them, the more accurate they become.

 AI achieves incredible accuracy through deep neural networks – which was previously
impossible. For example, your interactions with Alexa, Google Search and Google Photos are all
based on deep learning – and they keep getting more accurate the more we use them. In the
medical field, AI techniques from deep learning, image classification and object recognition can
now be used to find cancer on MRIs with the same accuracy as highly trained radiologists.

 AI gets the most out of data. When algorithms are self-learning, the data itself can become
intellectual property. The answers are in the data; you just have to apply AI to get them out.
Since the role of the data is now more important than ever before, it can create a competitive
advantage. If you have the best data in a competitive industry, even if everyone is applying
similar techniques, the best data will win.

10
 USES OF AI

Every industry has a high demand for AI capabilities – especially question answering systems
that can be used for legal assistance, patent searches, risk notification and medical research.
Other uses of AI include:

Health Care

AI applications can provide personalized medicine and X-ray readings. Personal health care
assistants can act as life coaches, reminding you to take your pills, exercise or eat healthier.

Retail

AI provides virtual shopping capabilities that offer personalized recommendations and discuss
purchase options with the consumer. Stock management and site layout technologies will also be
improved with AI.

Manufacturing

AI can analyze factory IoT data as it streams from connected equipment to forecast expected
load and demand using recurrent networks, a specific type of deep learning network used with
sequence data.

Banking

Artificial Intelligence enhances the speed, precision and effectiveness of human efforts. In
financial institutions, AI techniques can be used to identify which transactions are likely to be
fraudulent, adopt fast and accurate credit scoring, as well as automate manually intense data
management tasks.

11
 Introduction to Reinforcement Learning
There are three general categories of learning that artificial intelligence (AI)/machine learning
utilizes to actually learn. They are Supervised Learning, Unsupervised Learning and
Reinforcement learning

 Supervised Learning: The machine has a “teacher” who guides it by providing sample inputs
along with the desired output. The machine then maps the inputs and the outputs. This is similar
to how we teach very young children with picture books.

 Unsupervised Learning: This is the most important and most difficult type of learning and
would be better titled Predictive Learning. In this case the machine is not given any labels for its
inputs and needs to “figure out” the structure on its own. This is similar to how babies learn early
in life. For example they learn that if an object in space is not supported it will fall.

 Reinforcement Learning: Reinforcement learning is an area of Machine Learning. It is about

taking suitable action to maximize reward in a particular situation. It is employed by various
software and machines to find the best possible behavior or path it should take in a specific
situation. Reinforcement learning differs from the supervised learning in a way that in supervised
learning the training data has the answer key with it so the model is trained with the correct
answer itself whereas in reinforcement learning, there is no answer but the reinforcement agent
decides what to do to perform the given task. In the absence of a training dataset, it is bound to
learn from its experience.

Main points in Reinforcement learning –

 Input: The input should be an initial state from which the model will start
 Output: There are many possible output as there are variety of solution to a particular
problem
 Training: The training is based upon the input, The model will return a state and the user
will decide to reward or punish the model based on its output.
 The model keeps continues to learn.
 The best solution is decided based on the maximum reward.

12
Here we are considering an environment
(maze) which is to be completed by our
agent. And the maze has a fire pit and a
final end point.

Bellman Equation

Concept

s- State a-Action R-Reward ᵧ-Discount

V-Value

Here the maximum reward for completing the maze is +1 and falling in fire pit give -1 as reward.
Now we mark the values for each block as the agent would do while exploring the maze. So
considering it might take a random route like us and
mark the blocks +1 which lead it towards the goal. But
the problem arises when it reaches to top left corner. It
no more knows how to proceed further. So we need a
way to solve this problem.

This where Bellman equation comes in play.

Now takin same maze and marking the values of each block in
backward manner. We again consider same rewards as before
i.e. “+1” for completion and “-1” for falling in fire pit. In
addition to this we are now considering discounting factor to
be “-0.9” . Now using Bellman equation we can see that our
problem is solved as each block is now having different value
and the value increases as the agent is going towards the goal.

13
 Following the same equation we can calculate values for
every block in the maze

Advanced version of Bellman equation

We can clearly notice that there is somethings odd with the values of blocks that are connected to
the one with value “0.86” have different values.

This because we are now considering randomness which is another factor involve in our maze.
There is no way we can say that something is definitely going to happen we just take
probabilities of an outcome.

So there is always a chance that agent tries to move upwards but due to various factors which are
not in our control the agent end up in right or left direction.

This is the reason why we get a lower value for the block adjacent to the fire pit even though it
also leads us the block with a maximum value of +0.86.

This advanced form is also considering that randomness .

14
Markov Decision Process (MDP)

In order to understand MDP let’s understand what is Deterministic and Non-Deterministic

Search is.

Deterministic And Non-Deterministic Search

S.no Deterministic Search Non- Deterministic Search

1. In here there is no Randomness. In here there is some Randomness.

2. The probability of any action is 100%. The probability of any action is not 100%.

EXAMPLE

If agent in an environment tries to GO UP If agent in an environment tries to go up it

3 may go LEFT (10%) or UP (80%) or
it goes UP.
RIGHT (10%).

Markov Process is a process where the result depends only on the current state and not on past
action.

15
Markov Decision Process is also similar thing but here the results contain some randomness.

A Markov Decision Process (MDP) model contains:

 A set of possible world states S.

 A set of Models.
 A set of possible actions A.
 A real valued reward function R(s,a).
 A policy the solution of Markov Decision Process.

What is a Model?

 A Model (sometimes called Transition Model) gives an action’s effect in a state. In

particular, T(S, a, S’) defines a transition T where being in state S and taking an action ‘a’
takes us to state S’ (S and S’ may be same).

 A State is a set of tokens that represent every state that the agent can be in.

 An Action A is set of all possible actions. A(s) defines the set of actions that can be taken
being in state S.

 A Reward is a real-valued reward function. R(s) indicates the reward for simply being in
the state S

16
Plan and Policy

Plan - Plan in Artificial Intelligence is about the decision

making tasks performed by the robots or computer programs
to achieve a specific goal. The execution of plan is about
choosing a sequence of actions with a high likelihood to
complete the specific task.

Policy - As soon as randomness comes in picture we can’t

have a predefined plan as due to the randomness in the
environment there is always a possibility of getting into a state
that was not intended in the plan.

For example:- An agent in an environment with randomness tries to move forward but ends up in
left direction.

Living Penalty

It can be considered as a cost for staying inside the

environment.

The agent it gains negative reward for every action the

longer it stays in the environment more negative reward is
accumulated.

17
This reward is a very small value but it makes the agent to complete the
environment as soon as possible.

In the given example this can be seen clearly the agent tries to
complete the maze as soon as possible when there is a living
penalty.

As the reward for each step taken is -0.04 the agent tries to go up even when
there is a possibility of ending in fire pit.

Deep Learning
Deep learning is an artificial intelligence (AI) function that imitates the workings of the human
brain in processing data and creating patterns for use in decision making. Deep learning is a
subset of machine learning in artificial intelligence that has networks capable of learning
unsupervised from data that is unstructured or unlabeled. Also known as deep neural learning or
deep neural network.

 Deep learning is an AI function that mimics the workings of the human brain in
processing data for use in detecting objects, recognizing speech, translating languages,
and making decisions.
 Deep learning AI is able to learn without human supervision, drawing from data that is
both unstructured and unlabeled.
 Deep learning, a form of machine learning, can be used to help detect fraud or money
laundering, among other functions.

Deep learning unravels huge amounts of unstructured data

that would normally take humans decades to understand
and process.

The Neuron
The neuron is the basic working unit of the brain, a specialized
cell designed to transmit information to other nerve cells,
muscle, or gland cells.

The image shows a simplified diagram for human neuron and is

18
connection with other neuron.

The Dendrites receives information and the terminal passes that information to the other neuron
to which it is connected.

Similar is the case with Deep learning. Here we are trying to create an artificial neuron that

can mimic the function that a human neuron does.

Every Neural Network

has an input layer and
output layer.
There may or may not
be hidden layer.

The neural network

which we have taken
here gives us the price
for a house in an area.
Working of Neuron
It takes input from
input layer and gives
Simple Artificial Neuron
an output which in our
case is a weighted sum Where W1, W2, W3, W4 represents weights assigned to each
neuron.
of input values.

We have considered As our artificial neuron takes gets these values it assigns
some important factors different weights to each and every arm. If the area is good,
close to a distance, Age of building is not much and is in good
which can help us to
condition all there variables gets some weights.
get the price

In the end everything depends on our algorithm in use. Because

19
sometimes when a building is more than 100year old its cost increases as it is considered as a
part of history and shows how structures looks in those days.

How Neural Network Learn

Now let’s consider the hidden layer of neuron

which is the reason why neural networks are
able to give accurate answers to a problem.

For instance we are considering only 3 neurons

from our hidden layer.

The point to remember here is that all neurons

are connected to all the neuron in input layer, but we are only considering only specific neurons
which are triggering a particular neuron in our hidden layer.

First neuron is only considering the Area and

the Distance from cities. As the distance of land from city increases the cost of that particular
land decreases. Hence this particular neuron is searching for a larger area which is closer to city.

The Activation Function

An activation function is a function that is

added into an artificial neural network in order to
help the network learn complex patterns in the
data. When comparing with a neuron-based
model that is in our brains, the activation
function is at the end deciding what is to be fired
to the next neuron.

20
Some mostly used activation function

1) Unit step (threshold)

The output is set at one of two levels,

depending on whether the total input is
greater than or less than some threshold
value.

2) Sigmoid

The sigmoid function consists of 2

functions, logistic and tangential. The values
of logistic function range from 0 and 1 and -1
to +1 for tangential function.

21
3) Tangens hyperbolicus: Tanh

It works similar to the Sigmoid function, but has some

differences.
First, the change in output accelerates close to x=0,
which is similar with the Sigmoid function.
It does also share its asymptotic properties with Sigmoid:
although for very large values of x the function
approaches 1, it never actually equals it.
On the lower side of the domain, however, we see a
difference in the range: rather than approaching 0 as
minimum value, it approaches −1.

Rectified Linear Unit: ReLU

In order to improve on these observations, another activation

was introduced. This activation function, named Rectified
Linear Unit or ReLU, is the de facto first choice for most deep
learning projects today. It is much less sensitive to the problems
mentioned above and hence improves the training process.

And can be represented as follows:

f(x)={0,x,if x<0otherwise
Or, in plain English, it produces a zero output for all inputs smaller than zero; and x for all other
inputs. Hence, for all inputs<=0, it produces zero outputs.

Gradient Descent
Gradient descent is a first-order iterative
optimization algorithm for finding a local
minimum of a differentiable function

22
It is used to adjust the weights in a neural
network.

Here all the data is input provided to the

Network first and then then it performs
operations in order to optimize weights.

Simple but not so practical as it may result in

local maxima and not the global maxima due to
the curse of dimensionality ( the shape of the
curve changes).

This problem is solved by using Stochastic

Gradient Descent

Stochastic Gradient Descent

The word ‘stochastic‘ means a system or a

process that is linked with a random probability.
Hence, in Stochastic Gradient Descent, a few
samples are selected randomly instead of the
whole data set for each iteration.

23
BackPropagation

Forward Propagation
The training data is fed into the network it propagates forward and we get some output.

Then the errors are calculated which are fed back in the network in opposite direction and the weights are
adjusted in such a way that the network when again interprets the output it has less error than the previous
tries

Back propagation algorithm has very sophisticated mathematics which allows the neural networks to
adjust all the weights simultaneously.

24
Module in Course
https://drive.google.com/drive/folders/1J2S5kk18sGLye3azDmhPBIfwSXQxf2Qa?usp=sharing

Self-Driving car

25
26
 Software & Language

Computer programming languages allow us to give instructions to a computer in a language the

computer understands. Just as many human-based languages exist, there are an array of computer
programming languages that programmers can use to communicate with a computer. The portion
of the language that a computer can understand is called a “binary.” Translating programming
language into binary is known as “compiling.” Each language, from C Language to Python, has
its own distinct features, though many times there are commonalities between programming
languages.

These languages allow computers to quickly and efficiently process large and complex swaths of
information. For example, if a person is given a list of randomized numbers ranging from one to
ten thousand and is asked to place them in ascending order, chances are that it will take a sizable
amount of time and include some errors.

There are dozens of programming languages used in the industry today. some of these languages
are listed below.

C, C++, Java, JavaScript,Python

Python

Python is an advanced programming language that is

interpreted, object-oriented and built on flexible and robust
semantics. Python was developed in the late 1980s at CWI
in the Netherlands and first released to the public in 1991.

27
WHO USES IT?

 Professions and Industries:

o Python developers, software engineers, back end developers, Python
programmers
o Used by employers in information technology, engineering, professional
services and design
 Major Organizations: Google, Pinterest, Instagram, YouTube, DropBox,
NASA, ESRI
 Specializations and Industries: Web and Internet development (frameworks,
micro-frameworks and advanced content management systems); scientific and
numeric computing; desktop graphical user interfaces (GUIs)

WHAT MAKES LEARNING IT IMPORTANT?

Python lets you work quickly to integrate systems as a scripting or glue language. It’s
also suited for Rapid Application Develop (RAD).

 The game Civilization 4 has all its inner logic, including AI, implemented in
Python.
 NASA uses Python in its Integrated Planning System as a standard scripting
language.
 Features:
o Simple to learn and easily read
o Associated web frameworks for developing web-based applications
o Free interpreter and standard library available in source or binary on
major platforms
Python was developed in the late 1980s at CWI in the Netherlands and first released
to the public in 1991.

28
 ANACONDA IDE
Anaconda is a free and open-source distribution of the Python and R programming languages
for scientific computing (data science, machine learning applications, large-scale data
processing, predictive analytics, etc.), that aims to simplify package management and deployment.
The distribution includes data-science packages suitable for Windows, Linux, and macOS. It is
developed and maintained by Anaconda, Inc., which was founded by Peter Wang and Travis
Oliphant in 2012. As an Anaconda, Inc. product, it is also known as Anaconda
Distribution or Anaconda Individual Edition, while other products from the company are
Anaconda Team Edition and Anaconda Enterprise Edition, which are both not free.

Spyder

Spyder is an open source cross-platform integrated development

environment (IDE) for scientific programming in the Python language.
Spyder integrates with a number of prominent packages in the
scientific Python stack,including NumPy, SciPy, Matplotlib, pandas,
IPython, SymPy and Cython, as well as other open source software.It
is released under the MIT license.

29