UNIT I INTRODUCTION

Introduction to AI:

Artificial intelligence - Mimicking the behavior of set of rules to make decision based on the
thinking capabilities as human for a problem in either existence or nonexistence.

Artificial intelligence - Applying science and engineering to solve a problem by the computation
intelligence using machines.

Involving the machines to do a task with respect to the problem, that can be understandable as
living things including humans.

AI denoted by Alan Turing’s seminal work published in 1950 titled as “Computing machinery and
intelligence”.

Alan Turing – Father of theoretical computer science – asked – Can machines think?

How humans and computers thinking? What about the response from both?

Turing test –

Computing machinery and intelligence

“Can machines think?”.

“Imitation game” played by a man (A), a woman (B) and an interrogator (C) who stays away from
other two and may be either of sex. i.e., interrogator can’t able to see others.

Object of imitation game is that, the interrogator has to identify who are A and B and can be
labelled as X and Y respectively. At the end of the game, integrator must answer that, A should be
either X and Y and vice versa. Interrogator plays the game with some questions.

Interrogator: Will X please tell me the length of his or her hair? Answer will be collected from
both A and B. Teleprinter were used for the ideal communication between A, B and interrogator.
Human can answer to any questions from different perspective with respect to the previous
experience or knowledge, because of their thinking and acting capability.
What will happen in the system when a machine takes the role of A? Questions asked earlier will
be asking now to a machine and a human by the interrogator. Responses could decides the whether
the machines thinking or not. (Detail it with necessary sketches)

Python – founder – Guido van Rossum (Dutch programmer) – simple and convenient – Monty
Python’s flying circus – ease of use, readability, efficient, highly productive

python.org

Approaches in AI:

Human approach: 1. System that think like humans, 2. System that act like human

Ideal approach: 1. System that think rationally, 2. System that act rationally

Turing defined that, AI as system that act like humans which is categorized in the human approach.

AI can be classified into machine learning (ML) and deep learning (DL). So, ML and DL are
conjunction with AI. These leads to create an environment to make predictions through better
algorithms for real life problems based on input data.

Types of AI: 1. Weak AI, 2. Strong AI

Weak AI can be defined as the AI that trained and focused to perform specific tasks and is also
called as Narrow AI or Artificial Narrow Intelligence (ANI). It is a more accurate descriptor. Eg:
Apple’s Siri, Amazon’s Alexa, autonomous vehicles and so on.

Combining AGI and ASI together will develops Strong AI. Strong AI is still theoretical, and the
developments are expected.

Artificial General Intelligence (AGI) or General AI is a theoretical form in which the machine
intelligence would be same as human. i.e., Self-aware consciousness to solve the problems by the
consequent learning leads the futuristic plans.

Artificial Super Intelligence (ASI) known as superintelligence could surpass the intelligence and
ability of the human brain. Machines thinking beyond the human intelligence may be defined as
ASI. Eg: Science fiction movies
Terms machine learning and deep learning are interchangeable but has a vast difference.

Machine learning and deep learning are the sub-fields of AI, since the deep learning is a subfield
of machine learning.

Neural networks (Please visualize the critical path method) is a subset of deep learning and
consists of more than three layers including input and output also (Deep learning algorithm).

Machine learning is a branch of AI and computer science that, focuses on the use of data and
algorithms to imitate the way of human learning with a gradual increase in precision of the results.
It is important for the data science growth.
Arthur Samuel coined the term “machine learning”. Algorithms could be trained to predict the
problem statements using static methods in machine learning.

Deep learning and machine learning varies based on how the algorithm learns from the problem
formulation. Deep learning automates the analysis in the problem solving whereas some of the
human interventions can be avoided even with the large set of data.

Deep learning can be further defined as scalable machine learning.

Classical or non-deep machine learning is more dependent on the human interventions to learn.
So, human experts listing the hierarchy of the data set features to understand and learn about
structured data.

Deep learning known as supervised learning automatically evaluates the feature of the unstructured
data which understands and learns from the data given. Also, it requires very minimal or no human
intervention during the analysis of the problem.

Applications of AI:

Speech recognition [using natural language processing (NLP)] – Eg: Google assistant

Customer service – Online chatbots (ecommerce, messaging, etc)

Computer vision – image recognition (Facebook, insta – identification of face)

Searching engines – Eg: Google

And so on.

History of artificial intelligence: (Timeline)

1950 – Alan Turing’s – Turing test – World War II – Can machines think – denoted in Computing
machinery and intelligence.

1956 – John McCarthy – coined – 1st AI conference.

1980 – backpropagation algorithm was used to train the model.

1997 – IBM’s deep blue beats then world chess champion Garry Kasparov
 UNIT 2 MACHINE LEARNING – THE FUNDAMENTALS

Machine learning

Machine learning can be defined as that, “Machines thinks and acts from previous learning through
training using data sets”.

Machine learning is a subset of AI that focuses on learning from its experience to predict the
solution of a problem.

Working principle of machine learning:

Trained data sets are used in the machine learning algorithms to generate a model, that offers
appropriate yield/ results (validation). Simulation of a new data will be carried out based on the
validation to predict the solution. Further, the deployment of the model will be decided based on
the solution accuracy received from the analysis.

Types of machine learning:

1.Supervised learning, 2.Unsupervised learning, 3. Reinforcement learning

Supervised learning is always monitored by the humans in which the data sets will be trained for
the analysis. Later, the prediction could be possible.

Methods: Regression (linear and polynomial), Decision tree, random forest, classification (K-
Nearest neighbour, Trees, Logistic regression, Naïve-Bayes, Support vector machine)

Unsupervised learning is a method in which through observation and finding (patterns and
relationships) were happens automatically from the relationship using data clusters. Eg: Computer
vision

Methods: Clustering [Singular value decomposition (SVD), principal component analysis (PCA),
K-means], association analysis (Apriori, frequent pattern (FP) – Growth), Hidden Markov model
Reinforcement learning is the ability of the system generated from trained data set interacts with
the environment and offers the best solution using hit and trail method. Marks as rewards, that can
be added to the system based on its response and vice versa. Finally, trained data set will be ready
to predict the solution of a new data.

Supervised learning

Supervised learning or supervised machine learning is a subfield of machine learning and is

Eg: Spam classification in a separate folder

In supervised learning, Datasets involved in the analysis to train the model until getting desired/
appropriate outcome. Here, training data sets are inclusive of input and correct output, allows the
model to study/ learn over the time.

Supervised learning can be classified into classification and regression when solving data mining.

Classification algorithm used to accurately assign the data into specific categories. It recognizes
specific entities within the dataset and attempts to make some conclusion on how the entities
should be labelled or defined. Common classification algorithms are linear classifiers, support
vector machine (SVM), decision trees, k-nearest neighbor, and random forest.

Regression is employed to study the relation between dependent and independent variables. Eg:
Projections in sales/ business.

Types of regression: Linear regression, logistical regression, polynomial regression and so on.

Linear regression - Dependent variable are continuous

Logistical regression – dependent variable is categorical – Binary outputs (True/ false) – used to
solve binary classification problems. Eg: spam identification
Polynomial regression –

SVM- popular supervised learning – Developed by Vladimir Vapnik – used for both data
classification and regression. It is typically leveraged for classification problems, constructing a
hyperplane where the distance between two classes of data points is at its maximum. Hyperplane
can be known as the decision boundary, separating the classes of data points on either side of the
plane.

K-nearest neighbor/ KNN algorithm is a non-parametric algorithm, classifies data points based on
their proximity and association to other available data. It assumes that, similar data points can be
found near each other. It search to calculate the distance between data points (Euclidean distance).
It assigns a category based on the most frequent category or average. KNN is used in engines
(search) and image recognition.

Random forest used for both classification and regression purposes. “Forest” refers a collection of
uncorrelated decision trees which are merged to reduce variance and generate more accurate data
predictions.

Supervised learning examples

Image and object recognition (Computer vision) – locate, isolate and categorize objects out of
videos/ images

Predictive analysis – creating predictive analytics system to provide deep insights into various
business data points (Anticipation of business trend)

Customer sentiment analysis – context, emotion and intent with very little human intervention
(Customer interaction)

Spam detection – recognizes the pattern or anomalies through trained datasets, spam or non-spam
organization.

Challenges of supervised learning

- Requires expertise for a structure accurately