Module-1
Introduction to AI
Topic to be Covered
• AI vs ML vs DL
• Applications & business use cases
AI vs. ML vs. DL
What is Artificial Intelligence (AI)?
• Artificial Intelligence is a branch of computer science that enables machines to simulate human intelligence.
• General Characteristics of AI systems are reasoning, learning, problem solving, understanding natural language, and
making decisions.
• AI can be broadly divided into 2 types based on “how intelligence is incorporated into the machines?” or “how AI is
achieved?”
• Traditional AI: Earlier AI Systems, are rule-based, hence known as “rule-based-systems”, and “expert systems”
(complex and better than rule-based-systems). These systems had a set of explicitly defined rules and made
decisions based on these rules.
• For example, if condition A is met, then action B is taken.
• Modern AI: In modern times, AI is achieved or implemented using two most popular technologies “Machine
Learning” and “Deep Learning”.
• It can be said that ML and DL are the tools used to achieve AI.
Applications of AI:
•Autonomous Vehicles: Self-driving cars, like those developed by Tesla, use AI to make real-time decisions on the road
based on data from sensors and cameras.
•AI Chatbots: Virtual assistants like Apple’s Siri or Amazon’s Alexa use AI to understand and respond to user commands.
Also, ChatGPT and Perplexity are advanced chatbots powered by AI.
What is Machine Learning (ML)?
• Machine Learning is a subset of AI where computers learn patterns
Collect
from historical data instead of being explicitly programmed. Data
• ML models learn on their own from the data, it is provided and also
improve through exposure to data. Preprocess Data
• Traditional programming involves writing explicit code for specific
tasks. In ML, the model “learns” the rules or patterns from the Train Model
data it processes, rather than having them pre-defined.
• Basically, ML teaches the computer through algorithms to solve Test Model
problems by looking at hundreds or thousands of examples,
learning from them, and then using that experience to solve the same Predict
problem in new situations. Output
Types of ML
1. Supervised learning
• Supervised learning trains on labeled data with known outcomes.
• In supervised learning, the algorithm is provided with a training dataset consisting of input features
and corresponding correct output labels. The goal is to learn a mapping rule so the model can
accurately predict outcomes for new, unseen data.
Common tasks include:
Age Estimated Purchased
Salary
Classification: Predicting categorical labels (e.g., classifying whether a customer will purchase or not a product). 25 30000 0
30 45000 0
35 60000 1
Regression: Predicting continuous numerical values (e.g., forecasting house prices based on square footage and location).
Area Bedrooms Price
1000 2 200000
1200 2 230000
Examples: Common Algorithms: 1500 3 280000
•Email spam detection •Linear Regression
•Disease classification •Decision Tree
•House price prediction •Random Forest
•SVM
2. Unsupervised Learning
• Unsupervised learning algorithms work with completely unlabeled data.
• The model operates without a "teacher" or predefined answers, and instead
autonomously groups, organizes, or simplifies the raw data to discover hidden pattern.
Common tasks include:
•Clustering: Grouping similar data points together (e.g., segmenting retail customers into distinct purchasing profiles).
•Association: Finding relationships between variables (e.g., identifying that buyers of product X also frequently buy
product Y).
Examples: Common Algorithms
•Customer segmentation •K-Means
•Market basket analysis •Hierarchical Clustering
•Recommendation grouping •PCA
3. Reinforcement Learning
• Reinforcement learning (RL) does not rely on static datasets.
• Instead, an AI "agent" learns to make decisions by performing actions in a specific environment and receiving
feedback in the form of rewards or penalties.
• Through continuous trial and error, the agent figures out the sequence of actions that maximizes its cumulative
reward over time.
Common application includes:
•Game AI: Mastering complex environments like Chess, Go, or video games by playing against itself.
Examples:
•Robot navigation
•AlphaGo
•Autonomous vehicles
•Game AI
What is Deep Learning (DL)?
• Deep Learning is a subset of Machine Learning that uses Artificial Neural
Networks with many hidden layers.
• Unlike ML, Deep Learning automatically extracts important features from raw
data.
• Hence, DL is a subset of ML that deals with more complex problems, and huge
data and it is inspired by “how our brain works” i.e., It mimics the working of
the human brain.
• DL is inspired by Biological Neural Networks (in brain) and is based on Artificial
Neural Networks which have the capability of solving more complex problems or
tasks that ML is unable to solve.
1. Biological Neurons
Dendrite: It receives signal from other neurons
Synapse: Point of connection
Soma: Information processing
Axon: Transmits information
Biological Neuron Vs Artificial Neuron
Biological Neuron
The Perceptron(Artificial Neuron or Neural Network Unit)
What is Perceptron?
A perceptron is a basic unit in neural networks, mimicking brain neurons. It processes inputs with weighted connections
and a bias, producing binary outputs through an activation function. Primarily it is used in binary classification tasks.
Main Components of Perceptron
Rosenblatt’s perceptron is basically a binary classifier. The perceptron consists of 3 main parts:
•Input nodes or input layer: The input layer takes the initial data into the system for further processing. Each input
node is associated with a numerical value. It can take any real value.
•Weights and bias: Weight parameters represent the strength of the connection between units. Higher is the weight,
stronger is the influence of the associated input neuron to decide the output. Bias plays the same as the intercept in a
linear equation.
•Activation function: The activation function determines whether the neuron will fire or not. At its simplest, the
activation function is a step function, but based on the scenario, different activation functions can be used.
Example :
Locality pointer Price
Output Layer
Hidden layers are
Given inputs are derived from
the attributes some attributes
(Input Layer)
The Perceptron(Artificial Neuron or Neural Network Unit)
Single Layer Neural Network
b1
Input Layer Hidden Layer
𝐿𝑒𝑡 𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 𝑜𝑢𝑡𝑝𝑢𝑡
x1 w1
𝑌 ^=1
Step-1 Step-2 Output
x2 w2
Step-1 (Summation of weights along with inputs and bias)
Z=w1.x1+w2.x2+b
𝑍 = σ𝑛𝑖=1 𝑤𝑖 𝑥𝑖 + 𝑏 𝑠𝑖𝑚𝑖𝑙𝑎𝑟 𝑡𝑜 𝑦 = 𝑚𝑥 + 𝑐
Step-2 (Activation Function) [for non linearly separable problem] (Activation function is a function used for transforming the values
between 0 to 1)
A biological neuron only fires when a certain threshold is exceeded. Similarly, the artificial neuron will also only fire when the sum of the
inputs (weighted sum) exceeds a certain threshold value, let’s say 0.
If ∑wi*xi + b > 0:
output = 1
else:
Step-3 Calculate error output = 0
Loss = [Y- 𝑌 ^]=error
2. ANN
How does Deep Learning (DL) work? (Cont…)
DL involves a deeper level of automation in comparison with ML Algorithms:
[Link] Engineering: ML algorithms require feature extraction which is quite complicated. For example, if you want to
determine whether a particular image is of a cat or dog using the ML model. You have to manually extract features from the
image such as size, color, shape, etc., and then give these features to the ML model to identify whether the image is of a dog
or cat. However, DL models do not require any feature extraction and image itself can be provided as an input and they’re
capable of extracting features on their own.
[Link] Data: The accuracy of ML models stops increasing with an increasing amount of data after a point while the accuracy
of the DL model keeps on increasing with increasing data.
Raw Data
Data cleaning
Feature Engineering
Feature Selection
Feature Extraction
Feature Transformation
Feature Scaling
ML/DL Model
What is Deep Learning (DL)? (Cont…)
Applications of DL include:
•Facial Recognition: Facebook and Google Photos use deep learning to identify faces in images automatically.
•Also, Snapchat filters use DL to accurately detect and apply effects directly to specific facial features like the eyes, nose, and
mouth.
•Image Recognition: Self-checkout machines in grocery stores can use DL for product recognition, even without barcodes.
Some Popular Deep Learning Models are:
•CNN (Image Recognition)
•RNN (Time Series)
•LSTM (Speech Recognition)
•Transformer (ChatGPT, Gemini)
Applications
Applications of Artificial Intelligence (AI)
Application Example
Siri, Alexa, Google
Virtual Assistants (Understand voice commands and perform tasks)
Assistant
Chatbots (Provide automated customer support) Bank chatbots, ChatGPT
Warehouse robots,
Robotics (Perform repetitive and dangerous tasks)
industrial robots
Medical diagnosis
Expert Systems (Assist in decision-making)
systems
Google Nest, Amazon
Smart Homes (Control home appliances automatically)
Echo
Autonomous Vehicles (Make intelligent driving decisions)
Tesla Autopilot
AI-based intrusion
Cybersecurity (Detect and respond to cyber threats)
detection
Healthcare (Hospital management and virtual consultation) AI health assistants
Applications
Applications of Machine Learning (ML)
Application Example
Recommendation Systems (Suggest products, movies, or songs) Amazon, Netflix, Spotify
Spam Email Detection (Classify emails as spam or not) Gmail
Fraud Detection (Detect suspicious financial transactions) PayPal, Visa
Credit Scoring (Predict loan eligibility) Banks
Demand Forecasting (Predict future product demand) Walmart, Amazon
Stock Market Prediction (Analyse trends for investment) Financial institutions
Customer Segmentation (Group customers based on behavior) E-commerce platforms
Disease Prediction (Predict diseases from patient data) Diabetes prediction
Applications of Deep Learning(DL)
Applications
Application Example
Image Recognition (Identify objects and people in images) Google Photos
Face Recognition (Authenticate users using facial features) Face Unlock on smartphones
Speech Recognition (Convert speech into text) Google Assistant, Siri
Natural Language Processing (Understand and generate human language) ChatGPT, Gemini
Autonomous Driving (Detect lanes, vehicles, and pedestrians) Tesla
Medical Image Analysis (Detect diseases from MRI, CT, and X-rays) Cancer detection
Object Detection (Detect multiple objects in images/videos) CCTV surveillance
ChatGPT, DALL·E, GitHub
Generative AI (Generate text, images, code, and music)
Copilot
Business Use Cases of AI, ML, & DL in real time Scenarios
AI ML DL
In Amazon • Alexa voice assistant • Product recommendation •Product image recognition
• Warehouse robots • Demand forecasting •Voice recognition
•Fraud detection
In Tesla DL • Autonomous driving • Driver behaviour prediction • Object detection using
CNN
• Lane detection
• Traffic sign recognition
Google • Google Assistant • Search ranking • Google Translate
• Spam filtering • Google Photos
• Gemini AI
References:
• [1] E. Alpaydin, Introduction to Machine Learning, 4th ed. Cambridge,
MA, USA: MIT Press, 2020.
• [2] S. Marsland, Machine Learning: An Algorithmic Perspective, 2nd
ed. Boca Raton, FL, USA: CRC Press, 2015.
• [3] F. Chollet, Deep Learning with Python, 2nd ed. Shelter Island, NY,
USA: Manning Publications, 2021.
• [4] M. Mohri, A. Rostamizadeh, and A. Talwalkar, Foundations of
Machine Learning, 2nd ed. Cambridge, MA, USA: MIT Press, 2018.
• [5] T. Mitchell, Machine Learning. New York, NY, USA: McGraw-Hill,
1997.