FOUNDATIONS OF MACHINE LEARNING

WINTER 24-25, ECED, NIT CALICUT

 SESSION PLAN

• Introduction to Machine Learning (ML) and role of data

• Industry profiles
• ML applications
• Types of learning
• Sample application: House price prediction
• Role of Linear Algebra in ML
• Matrix rank and existance of solution
 What is Machine Learning?
• In the past, scientists like Galileo and Kepler designed experiments to make
observations and collected data

• Scientists then came up with laws that explain the observetions in terms of
collected data

• In medicine, cures for many diseases were found by collecting information

from patients and analyzing them for commonalities and differences

• We want to automate this process of going from data to knowledge

 What is Machine Learning? (contd.)
• We want to automate this process of going from data to knowledge

• This is essential in the digital age where data is widely generated, recorded
and freely available

• At the same time, general problems have been studied well and algorithms
based on causality have saturated to a great extend

• Once, if you were smart, you invented a new algorithm; now, if you are smart,
you find a source of data for your learning algorithms
 Machine Learning
• In machine learning, the aim is to have a computer program that learns

• Learning means getting better through experience

• “Better” implies a performance criterion that is optimized

• “Experience” implies data collected in the past—for example, in past trials

• As the program sees more data, its output leads to higher performance
according to the criterion that is to be optimized
 Machine Learning (contd.)
• In machine learning, the aim is to have a computer program that learns

• The learning program is thus a general template with modifiable parameters

• By assigning different values to these parameters, the program can change

its performance

• The learning algorithm adjusts the parameters of the template—which we call

a model—by optimizing a performance criterion defined on the data

• With examples, the parameters are updated slightly at each example, so that
the performance improves gradually in time.
Machine Learning vs Programming
 Industry profiles related to ML
• Domain Experts: Have domain knowledge, business insights, know what
data is important and where to find it, identify the real impact of a ML model

• Software Development Engineers (SDE): Well-versed in programming with

the ability to collect, store and use data through data pipelines, train machine
learning models and deploy them for customer service

• Data Scientists: Full stack experts on data handling, data cleaning, data
mining, model training, model deployment and data visualization

• Machine Learning Engineers: Develop, customize, and improve state of the

art machine learning models for specific applications
 Applications of ML in Industry
• Pattern recognition
– Face recognition, Speech recognition, Biometric identification
– Facial expressions, Emotions from speech
– Optical character recognition, Handwritten text
– Medical diagnosis from scans like cancer detection

• Anomaly detection
– Fraudulent credit card transactions, money transfers
– Spam detection for calls, texts, e-mails
– Intrusion detection from videos
– Unusual readings from sensors in nuclear powerplant
 Applications of ML in Industry
• Prediction
– Stock prices, Housing prices or rents
– Trajectory of a vehicle like car or aircraft
– Customer preferences in shopping
– Maintainence of machines, quality control and service

• Generation of patterns
– Text to speech/image/video generation
– Style modification in speech/image/video
– Synthetic sensor data for power plants, vehicle navigation
– Customized content generation
 Types of Learning: Supervised Learning
• Given training data and desired outputs (labels)
 Types of Learning: Semi-Supervised Learning
• Given training data and few desired outputs (labels)
 Types of Learning: Unsupervised Learning
• Given training data without desired outputs (labels)
 Types of Learning: Reinforcement Learning
• Rewards from sequence of events
 Sample Application: House price prediction
• A realtor wants to list a house for sale in a particular neighborhood
• What will be an appropriate price of the house?
• If you are given data: details of sold houses in neighborhood, can you predict?
 House price prediction: Role of Linear Algebra
• Data is given in the following form:
– House 1 with two bedrooms and 10 cents plot area sold for Rs 30,00,000.
– House 2 with three bedrooms and 11 cents plot area sold for Rs 50,00,000.
• Given your house with three bedrooms and 8 cents plot area, can you predict
an appropriate price?
 House price prediction: Role of Linear Algebra
• Data is given in the following form:
– House 1 with two bedrooms and 10 cents plot area sold for Rs 30,00,000.
– House 2 with three bedrooms and 11 cents plot area sold for Rs 50,00,000.
• Given your house with three bedrooms and 8 cents plot area, can you predict
an appropriate price?

• If you consider no of bedrooms as feature x1 and plot area as feature x2

– House 1 data: 2x1 + 10x2 = 30,00,000
– House 2 data: 3x1 + 11x2 = 50,00,000

• Can you predict the price for x1=3 and x2=8 (cents) ?
[2 10 [x1 = 30,00,000
3 11] x2] = 50,00,000
 Deterministic vs Stochastic environment
• Deterministic environment in AI is an environment setting where the outcome
of an action is determined by the current state and action
• In deterministic environment, there is no randomness involved
• Given initial conditions and actions, the environment will always produce the
same outcome (predictable)
• In real-world situations, not all scenarios follow a predictable path. Many
involve factors like uncertainty, randomness, or incomplete information.
• Stochastic Environment in AI is the environment in which the outcomes are
affected by randomness.
• Stochastic environment introduces an element of unpredictability, that can
arise from various sources such as random events, probabilistic transitions
between states or incomplete information about environment.
• In stochastic environments, AI systems must integrate probabilistic models or
methods to manage uncertainty successfully.