Module 1

Introduction
Copyright © 2018 McGraw Hill Education, All Rights Reserved.

PROPRIETARY MATERIAL © 2018 The McGraw Hill Education, Inc. All rights reserved. No part of this PowerPoint slide may be displayed, reproduced or distributed in any form or by any
means, without the prior written permission of the publisher, or used beyond the limited distribution to teachers and educators permitted by McGraw Hill for their individual course preparation.
If you are a student using this PowerPoint slide, you are using it without permission.
 Introduction
Two basic premises of machine learning:
• The real world is pervasively imprecise and uncertain.
• The precision and certainty carry a cost.

Guiding principle of machine learning:

• Exploit tolerance for imprecision, uncertainty and
partial truth.
• Achieve tractability, robustness, and low solution
costs.
 Definition
“A computer program is said to learn from experience
with respect to some class of tasks and performance
measure, if the performance at the tasks, as measured
by the performance measure, improves with
experience”
Features of a well-defined learning problem:
• The learning task
• The measure of performance
• The task experience
 Task experience
They are mainly of two types:
• Experimental data (examples, samples,
measurements, records, patterns or observations).
• Structured human knowledge (experience, expertise,
heuristics) expressed in linguistic form (IF-THEN rules).

Our focus would be on the first form of experience.

 The Learning Task
Important aspects of ‘learning from experience’:
Remembering and Adapting: Recognition of a situation
and remembering the past outcome and proceeding as
per the same

Generalizing: Past knowledge helpful in analyzing

similar situations

Machine learning = Remembering + Adapting +

Generalizing
Machine learning is about modification or adaptation of
the decisions and actions so that those actions get more
accurate with experience.

Machine learning, like animal learning, relies heavily in

the notion of similarity in its search for valuable
knowledge in data which would help it generalize for the
unseen situations.
 Types of learning tasks
Four different types of learning tasks are there in the real-
world:
1. Classification learning– categorize or classify examples
with respect to its input to different categories or classes
2. Clustering – similar observations are grouped together
3. Regression – numerical output is sought from observations
4. Association learning – any connection or link among
observations is sought irrespective of the class value
Our focus here is on the first three type of tasks. Errors in
learning is the measure of performance for
classification/regression.
 Model of Learning
Machine
• The output of an algorithm represents the learned
knowledge.
• This knowledge is in the form of a model based on
structural information in the data.
Deployment of model for decision-making:
• It will make predictions with respect to the assigned
task for observations not in the task experience.
• A good model is one which will generalize well to the
observations unseen by the machine during training.
Block diagrammatic representation of a learning
Machine
 Data Mining
• The collection of methods for extracting (predictive)
models from data has to be known as the field of
machine learning.
• Data mining is a part of machine learning which
mainly examines real-world applications.
• Research focused on commercial applications and
business issues of data analysis tend to gravitate
towards data mining.
• However, both the fields are concerned with the
analysis of data to find informative patterns.
Artificial Intelligence (AI)
• AI, like machine learning, is any computer program
that does something smart.
• All machine learning counts as AI, but not all AI counts
as machine learning.
• For example, rule-based expert systems, frame-based
expert systems, knowledge graphs, evolutionary
learning could be described by AI but none of them is
machine learning.
 Deep Learning
• Deep learning may be considered as a subfield of
machine learning.

• Deep neural networks are a set of algorithms setting

new records in accuracy for many important problems.

• Deep refers to multiple number of layers in a neural

network which makes it more complex and
complicated to interpret and understand but, it learns
features of the data in a hierarchy.
 Analytics
• Term analytics, commonly used in business
community for data-driven analysis for decision
making.
• A general term – applying various analysis techniques
to data to solve problems.
• Not a technology in itself, but group of tools used in
combination.
• Purpose – to gain information, analyze information,
predict different possible actions, quantify the effects
of possible future decisions on business optimization.
 Big Data Analytics
• Scaling to complex, extreme large datasets

• More in terms of volume, variety and velocity

• Probably the most debated current research issue

 Our focus
• Technology is changing
• Data mining exists on the shifting landscape.
• Not only is the old part being redefined, but new areas
of interest always loom.

Our focus here is on “foundations of machine learning”.

 Data Representation
Some setof samples over which various output functions
is defined.
1. Training experience in the form ofpatterns

2. Pattern specified by a fixed numberof

attributes/features

where each feature has a numerical value for the pattern.

The domain of is given by a set of its values.
Pattern visualized in -dimensional state space :
• – finite set of feature vectors for all the patterns.
• can be visualized as a region of the state space i.e. .
Note that is the representation of ; and is the
representation space.
For supervised learning problem:
• the decision attribute (output) y - a priori
• For multiclass classification (pattern recognition)
problems, the domain of y is the set .
• In classification problems, each pattern is associated
with an output
The output belongs to one-dimensional region Y of the
state space.
 Forms of Learning
SUPERVISED/DIRECTED LEARNING
Machine designed by exploiting a priori known
information in the form of training examples having
input vectors: and output to each vector: .
The ‘supervisor’ has :

The dataset is used for interpreting a model of the

system.
• Assumption: Machine can work on generalized data
having form
• Then for any input in , machine should work fine and
predict it well.
• However, choice of features or attributes fed to the
system would change the output
Two types of supervised learning
1. Classification (pattern recognition)
2. Regression (numeric prediction).
Classification
• input-output pair in data
• – input vector with n features ,
• – output which is discrete class .
Goal: Predict that which of the classes each new vector
belongs to

Regression
• input-output data
• – regressors or attributes
• – continuous numeric quantity
Goal: Fit a function on the input-output data to predict
numeric output values for new inputs.
 Regression
• input-output data
• – regressors or attributes
• – continuous numeric quantity
Goal: Fit a function on the input-output data to predict
numeric output values for new inputs.
UNSUPERVISED/UNDIRECTED LEARNING
• - output not available
• - set of feature vectors
• Goal- Unravel the underlying similarities in the data
and make groups or clusters based on feature vector