Applied Machine Learning

SS/SE ZG568

Dr. Sugata Ghosal

CSIS Off-Campus Faculty
BITS Pilani
Course Content
No Title of the Module References
M1 Introduction to Machine Learning T1: Chapter 1
1. What and Why R2
2. Applications of Machine Learning
3. Types of Machine Learning
4. Challenges in Machine Learning

M2 Big Picture: End-to-end Machine Learning T1: Chapter 2

2.1. Framing the ML Problem
2.2. Data Types, Pre-processing, Visualization and Analysis T2: Chapter 2-3

M3 Big Picture: End-to-end Machine Learning T1: Chapter 2-3

• Model Selection and Training T2: Chapter 2-3
• Evaluation
• Machine Learning Pipeline

M4 Linear Prediction Models T1: Chapter 4

• Linear Regression T2: Appendix D
• Gradient Descent and Variants
• Regularization
• Bias Vs. Variance
M5 Classification Models I T1: Chapter 4, 5
1. Logistic Regression
2. Support Vector Machines T2: Chapter 5, Section 3
3. Naïve Bayes
4. Comparative Analysis
Course Content
No Title of the Module References
M6 Classification Models II T1: Chapter 6-7
6.1 Decision Tree T2: Chapter 4, 5.6
6.2 Challenges with Decision tree
6.3 Ensembles
6.3.1. Bagging
6.3.2. Boosting
6.3.3. Random Forest
M7 Unsupervised Learning T1: Chapter 8-9
7.1. Dimensionality reduction and feature extraction
7.2 K-means Clustering T2: Chapter 8
7.4 EM Algorithm
7.5 Applications
M8 Neural Networks T1: Chapter 10-11
8.1. Perceptrons and Delta Rule
8.2. Multi-Layer Perceptrons and Backpropagation T2: Chapter 5.4
8.3. Deep Learning
M9 Deep Networks T1: Chapter 14, 15
9.1. Convolutional Neural Network
9.2. Recurrent Neural Network Lecture Slides
9.3. Applications in Image and Text Processing
M10 FAccT Machine Learning R1
10.1. Bias and Fairness
10.2. Interpretability and Transparency
10.3. Robustness and Adversarial Attacks
Text Books

T1 Aurelien Geron, “Hands-On Machine Learning with Scikit-Learn,

Keras and Tensorflow”, O’Reilly, 2020
T2 P-N Tan, M. Steinbach, Vipin Kumar, “Introduction to Data
Mining”, 2016, Pearson

R1 Christoph Molner, “Interpretable Machine Learning”, 2020,

https://christophm.github.io/interpretable-ml-book/
R2 Pedro Domingos, “A Few Useful Things to Know About Machine
Learning”, pp. 78-87, Communications of the ACM, vol. 55 no. 10,
October 2012.
Introduction to Machine
Learning

Dr. Sugata Ghosal

CSIS Off-Campus Faculty
BITS Pilani
In this segment

• What is Machine Learning

• Why Machine Learning
• Connection with Other technical areas
 What is Machine Learning (ML)?
What is Machine Learning

• The science (and art) of programming computers so they can learn from data

• More general definition

• Field of study that gives computers the ability to learn without being explicitly
programmed

• Engineering-oriented definition
• Algorithms that improve their performance P at some task T with experience E

• A well-defined learning task is given by <P, T, E>

Traditional programming Machine Learning

Data Data
Computer Output Computer Program
Program Output
 Defining the Learning Tasks
Improve on task T, with respect to performance metric P, based on experience
E
• Example 1
• T: Playing checkers
• P: Percentage of games won against an arbitrary opponent
• E: Playing practice games against itself
• Example 2
• T: Recognizing hand-written words
• P: Percentage of words correctly classified
• E: Database of human-labeled images of handwritten words
• Example 3
• T: Driving on four-lane highways using vision sensors
• P: Average distance traveled before a human-judged error
• E: A sequence of images and steering commands recorded while observing a
human driver.
• Example 4
• T: Categorize email messages as spam or legitimate.
• P: Percentage of email messages correctly classified.
• E: Database of emails, some with human-given labels
 Traditional Approach to Spam Filtering
Spam typically uses words or phrases such as “4U,” “credit card,” “free,” and “amazing”

• Solution
• Write a detection algorithm for frequently
appearing patterns in spams
• Test and update the detection rules until it is
good enough.

• Challenge
• Detection algorithm likely to be a long list of
complex rules
• hard to maintain.
 Machine Learning Approach
Automatically learns phrases that are good predictors of spam by detecting
unusually frequent patterns of words in spams compared to “ham”s

• The program is much shorter, easier to maintain, and most likely more
accurate.
 A Classic example of ML Task
It is very hard to say what makes a “2”

Actually a
“3” not a “2”
All “2”s

A “3”

A “7”
 More ML Usage Scenarios
Tasks that are best solved by using a learning algorithm
• Recognizing Patterns in images, text
• Facial identities or facial expressions
• Handwritten or spoken words
• Medical images

• Recognizing Anomalies in structured dataset

• Unusual credit card transactions
• Unusual patterns of sensor readings in a nuclear power plant

• Prediction from time series data

• Future stock prices or currency exchange rates

• Generating new Patterns

• Generating images or motion sequences
 Machine Learning Helps Human Learning
Sometimes reveals unsuspected correlations / new trends leading to better
understanding
• Data Mining
• Applying ML techniques to dig into large amounts of data can help discover
patterns
 When to Use Machine Learning
ML Usage Scenarios
• Problems for which existing solutions require a lot of hand-tuning or long lists of rules
• Complex problems for which there is no good solution at all using a traditional approach
• Human rationale cannot be codified
• Changing environments
• Get insights about complex problems and large amounts of data
Where does ML fit in?
 Applications

Dr. Sugata Ghosal

CSIS Off-Campus Faculty
BITS Pilani
In this segment

• Types of Applications
• Application Domains
• State of the Art Applications
Two Classes of Application

• Assign/Predict an object/event to an element of a given set

• If set of integers, then it is a classification problem
• If set of reals, then it is a regression problem
• Predict a sequence of steps to achieve a goal
Application Domains

• Internet
• Computational biology
• Finance
• E-commerce
• Space exploration
• Robotics
• Information extraction
• Social networks
• Software engineering
• System management
• Creative Arts
 Example: Classification
Assign object/event to one of a given finite set of categories
• Medical Diagnosis
• Credit card applications or transactions
• Fraud detection in e-commerce
• Worm detection in network packets
• Spam filtering in email
• Recommended articles in a newspaper
• Recommended books, movies, music, or jokes
• Financial investments
• DNA sequences
• Spoken words
• Handwritten letters
• Astronomical images
 Example: Planning, Control, Problem Solving

Performing actions in an environment in order to achieve a goal

• Playing checkers, chess, or backgammon
• Balancing a pole
• Driving a car or a jeep
• Flying a plane, helicopter, or rocket
• Controlling an elevator
• Controlling a character in a video game
• Controlling a mobile robot
Breakthrough in Automatic Speech Recognition

ML used to predict of phone states from the sound spectrogram

Deep learning has state-of-the-art

# Hidden Layers 1 2 4 8 10 12
results
Word Error Rate % 16.0 12.8 11.4 10.9 11.0 11.1

Baseline (Gaussian Mixture Model) performance =

15.4%
[Reference: Zeiler et al., “On rectified linear
units for speech recognition” ICASSP 2013]
Impact of Deep Learning in Speech
Technology
 Visual Question Answering
Answering questions about images
Autonomous Car Technology

Path
Planning

Laser Terrain
Mapping

Learning from Human

Adaptive
Drivers
Vision

Sebastian

Stanle
y
Contemporary ML Based Solutions

• Optical Character Recognition

• Product Image Classification on a production line
• Detecting tumors in brain scans
• Information extraction from document images
• Categorization of news articles
• Flagging offensive comments in discussion forums
• Document summarization
• Chatbots / Voice activated App
• Detecting Credit Card Fraud
• Customer Segmentation
• Recommendation System for products, movies, news
• Forecasting company revenue
Types of Machine Learning

Dr. Sugata Ghosal

CSIS Off-Campus Faculty
BITS Pilani
In this segment

• Supervised Vs. Unsupervised Vs. Semi-Supervised Vs. Reinforcement

• Batch Vs. Online Learning
• Instance Based Vs. Model Based Learning
 Types of Learning
Based on level of supervision
• Supervised (inductive) learning
• Given: training data, desired outputs (labels)
• Unsupervised learning
• Given: training data only (without desired outputs)
• Semi-supervised learning
• Given: training data and a few desired outputs
• Reinforcement learning
• Given: rewards from sequence of actions
Supervised Learning: Regression

• Given (x1, y1), (x2, y2), ..., (xn, yn)

• Learn a function f (x) to predict y given x
– y is real-valued

9
September Arctic Sea Ice
Extent (1,000,000 sq

8
7
6
5
4
3
2
km)

1
0
1970 1980 1990 2000 2010 2020
Year Number
Supervised Learning: Classification

• Given (x1, y1), (x2, y2), ..., (xn, yn)

• Learn a function f (x) to predict y given x
– y is categorical

Cancer (benign / malignant)

y=1 (malignant)

y=0 (benign)

Tumor Size (x)

Supervised Learning: Classification

• Given (x1, y1), (x2, y2), ..., (xn, yn)

• Learn a function f (x) to predict y given x
– y is categorical

Cancer (benign / malignant)

1 (malignant)

0 (benign)
Tumor Size Learnt classifer
If x>T, malignant else benign
Predict benign Predict malignant

x=T
Increasing Feature Dimension

• x can be multi-dimensional
– Each dimension corresponds to an attribute

- Clump Thickness
- Uniformity of Cell Size
Age
- Uniformity of Cell Shape
…

Tumor Size
Example: Supervised Learning
Techniques

• Linear Regression
• Logistic Regression
• Naïve Bayes Classifiers
• Support Vector Machines (SVMs)
• Decision Trees and Random Forests
• Neural networks
Unsupervised Learning

• Given x1, x2, ..., xn (without labels)

• Output hidden structure behind the x’s
• e.g., clustering
Example: Unsupervised Learning
Techniques

• Clustering
• k-Means
• Hierarchical Cluster Analysis
• Expectation Maximization
• Visualization and dimensionality reduction
• Principal Component Analysis (PCA)
• Kernel PCA
• Locally-Linear Embedding (LLE)
• t-distributed Stochastic Neighbor Embedding (t-SNE)
• Association rule learning
• Apriori
• Eclat
 Data Visualization
Visualize 2/3D representation of complex unlabelled training data
• Preserve as much structure as possible
• e.g., trying to keep separate clusters in the input space from overlapping in the
visualization
• Understand how the data is organized
• Identify unsuspected patterns.
frog

cat
bird
dog
truck

automobile

deer

horse

ship
airplane
 Applications: Unsupervised Learning
Genomics application: Group individuals by genetic similarity
Genes

Individuals
Applications: Unsupervised Learning
Organize computing clusters Social network analysis

Market segmentation Astronomical data analysis

 Semisupervised Learning
Partially labelled data – some labelled data and a lot of unlabelled data
• Combines unsupervised and supervised learning algorithms
• Photo hosting service, e.g., google photos
 Reinforcement Learning
A learning agent
• observes the state of the
environment, select and perform
actions
• gets +ve or -ve rewards in return
• learns the best strategy, aka a
policy, to get the most reward over
time.
• policy is a mapping from states →
actions

• Examples:
– Game playing, e.g., AlphaGo
– Robot in a maze
– Balance a pole on your hand
 Types of Learning
Based on how training data is used
• Batch learning
• Uses all available data at a time during training
• Mini Batch learning
• Uses a subset of available at a time during training
• Online (incremental) learning
• Uses single training data instance at a time during training
 Types of Learning
Based on how training data is used
• Instance Based Learning
• compare new data points to known data points
• Model Based learning
• detect patterns in the training data and build a predictive model

Instance Based Model Based