By Firoz Uddin

WHAT IS MACHINE LEARNING?

Machine Learning (ML) is a branch of artificial intelligence
(AI) that enables computers to learn from data and
improve their performance on tasks without being
explicitly programmed. Instead of writing code with fixed
rules, you feed the machine data, and it identifies
patterns to make predictions or decisions.
DOG DOG CAT
 lables

DOG DOG CAT

data
 Explanation:
Machine learning is like teaching a kid to learn things, just like how you learn from
experience. Imagine you're teaching a kid to recognize cats and dogs. You show it a
bunch of cats and dogs The kid doesn't know what a cat or dog is at first. But after
seeing lots of them, it starts to notice patterns:

Cats usually have pointy ears and whiskers.

Dogs are usually bigger and have different shapes.

Now, when you show the kid a new picture, it can guess if it's a cat or a dog because it
learned from the examples you showed it.
 What Does 'Learning' Mean in ML?
Learning in ML means:
The model improves its performance by finding patterns in data over time. It generalizes
from past examples to predict unseen data. The model adjusts its internal parameters (like
weights in neural networks) to minimize errors and improve accuracy.

Imagine teaching a child to recognize apples:

1. Show them pictures of apples and non-apples.
2. They start identifying features (round shape, red/green color).
3. Over time, they improve at recognizing apples without explicit instructions for every
type of apple.

Similarly, an ML model "learns" from data, refining its ability to identify patterns and make
accurate predictions.
AI - ML - DL
WHY IS MACHINE LEARNING?
Traditional Programming:
You write explicit rules to solve a problem.
Example: To check if a number is even, you code if (number % 2 == 0).

Machine Learning:
The machine finds the rules by analyzing data.
Example: Feed thousands of labeled images (cats vs. dogs), and the model learns to
classify them.
WHY DO YOU WANT TO LEARN ML?
TYPES OF MACHINE LEARNING
 1. Supervised Learning
In supervised learning, the model is trained on labeled data. The input data comes with
corresponding output labels, and the model learns to map the input to the correct output.
Goal: Predict outcomes based on input data.
Training Data: Labeled (contains input-output pairs).
How It Works:
a. The model is fed data with known labels.
b. It learns the relationship between inputs and outputs.
c. The model generalizes this relationship to predict outputs for new, unseen data.
Example Use Cases:
Classification: Spam detection (email labeled as spam or not spam).
Regression: Predicting house prices based on size, location, etc.
Examples of Algorithms:
Linear Regression
Decision Trees
Support Vector Machines (SVM)
Random Forest
 2. Unsupervised Learning
In unsupervised learning, the model is trained on data without labeled outputs. The goal is
to find patterns, structures, or relationships in the data.
Goal: Discover hidden patterns or groupings.
Training Data: Unlabeled (no output labels).
How It Works:
a. The model receives raw data.
b. It tries to find clusters, associations, or reduce dimensionality without explicit
labels.
Example Use Cases:
Clustering: Grouping customers by purchasing behavior.
Dimensionality Reduction: Reducing large datasets while retaining key information.
Anomaly Detection: Fraud detection by identifying outliers.
Examples of Algorithms:
K-Means Clustering
Principal Component Analysis (PCA)
Hierarchical Clustering
 3. Reinforcement Learning
In reinforcement learning (RL), the model (called an agent) learns by interacting with an
environment and receiving feedback in the form of rewards or penalties. Goal is to
maximize cumulative rewards over time.
Goal: Learn a sequence of actions that lead to maximum reward.
Training Data: No fixed dataset – the model learns through trial and error.
How It Works:
a. The agent takes actions in an environment.
b. Based on the outcome, it receives rewards (positive) or penalties (negative).
c. The model adjusts its strategy to maximize long-term rewards.
Example Use Cases:
Game Playing: AI mastering games like chess, Go, or video games.
Robotics: Teaching robots to walk or perform tasks.
Recommendation Systems: Learning user preferences dynamically over time.
Examples of Algorithms:
Q-Learning
Proximal Policy Optimization (PPO)
Actor-Critic Models
COMMONLY USED TERMS IN ML
1. Data is just information. For machine learning, it’s like the "fuel" that helps the
computer learn. This could be pictures, numbers, words, or anything the computer uses
to learn. Example: If you want to teach a computer to recognize dogs, the data could be
a bunch of pictures of dog.

2. Model is like a brain for the computer. It takes the data, learns from it, and tries to
make decisions or predictions. Example: If you show the model pictures of cats and
dogs, the model learns what makes a cat different from a dog, so it can tell the
difference later.

3. Training is like practice. The computer looks at lots of examples (data) and adjusts
itself to get better at making predictions. Example: If you’re learning to play the piano,
you practice a song many times. In machine learning, the computer practices by looking
at lots of data.
4. Algorithm is like a set of instructions or a recipe. It tells the computer how to learn
from the data. Example: If you're baking a cake, the recipe is the algorithm—it tells you
what to do step by step. In machine learning, algorithms help the computer "bake" a
good model!

5. Feature is a piece of information the computer uses to understand something. It's like
a clue that helps the model make a decision. Example: When recognizing a dog, features
could be things like its size, color, and shape of the ears.

6. Label is the answer or category that we want the computer to predict. It tells the
computer what the data should be. Example: In a picture of a dog, the label is “dog.” In
supervised learning, the computer learns to predict labels for new data
 7. Training Set and Test Set : The training set is the data the computer uses to learn,
while the test set is the data used to check how well the computer has learned.
Example: Imagine you have 10 pictures of cats and dogs. You show 8 of them to the
computer to help it learn (training set), and then you check if it can guess the labels of
the remaining 2 pictures (test set).

8. Overfitting happens when the computer learns the details of the training data too
well and becomes too specific. It can get confused when it sees new data because it’s
only good at recognizing the training examples. Example: It’s like memorizing a math test
and not understanding how to solve other problems. The computer can only do well on
the examples it saw before, not on new ones.
9. Underfitting happens when the computer doesn’t learn enough from the data. It’s like
not studying enough and failing to understand the problem. Example: If you give the
computer just one picture of a cat and one of a dog, it won’t learn enough to tell the
difference correctly on new pictures

10. Accuracy is how often the computer gets the right answer. The higher the accuracy,
the better it’s doing! Example: If the computer guesses “dog” 8 times out of 10 correctly,
its accuracy is 80%.

11. Hyperparameters are settings that control how a machine learning model learns. It’s
like choosing the right settings on your phone or video game to make sure it works just
right. Example: If you were training a computer to recognize dogs, hyperparameters
might control how fast it learns or how many times it looks at the data.
 Origin of Machine Learning
1950s - Early Al models focused on rule-based approaches.

1959 - Arthur Samuel coined the term "machine learning" while developing a
system that could learn to play checkers.

1980s-2000s - Development of neural networks and increased computational

power accelerated progress.

2010s to Present - Massive data availability, GPUs, and deep learning

revolutionized ML, enabling breakthroughs in fields like NLP, computer vision, and
robotics.
RECAP AND QUESTIONS