Introduction to DL

Introduction to Types of Neural Networks

Learning Outcome

we have already learned

Learned the Introduction to Deep Learning

Studied the Perceptron Model

Learned about Deep Learning frameworks:

• TensorFlow

• PyTorch

• Keras

Imagine different types of specialists in a hospital

Handles basic cases

 Eye Specialist                 Focused on vision

Treats memory issues

Creates new medicines

Research Lab

Memory Specialist

General Doctor

Each doctor has a specific role.

Neural Networks also work the same way. Different types of neural networks are designed for different kinds of tasks.

Some networks handle simple data

Some focus on images

Some understand sequences

Some remember long patterns

Some create new data

Just like specialist doctors solve different medical problems,
different neural networks solve different AI problems.

Neural Networks

Neural Networks are a type of Artificial Intelligence (AI) algorithm that work similar to the human brain. 

Neural Networks are:

Inspired by the Human Brain

Recognize Patterns

Used for Tasks

Processes Raw Data

 Feedforward Neural Network (FNN)

Feedforward Neural Network (FNN) is one of the simplest types of neural networks.

It is also called a Multi-Layer Perceptron (MLP).

Data moves in one direction Information flows only forward, not backward.

Used for prediction tasks

It is commonly used for classification (like spam detection) and regression (predicting values like price).

Convolutional Neural Network (CNN)

Key Points

Convolutional Neural Network (CNN) is a type of neural network mainly used to analyze images and videos.

1. Best for image and video data

2. Uses filters to detect patterns

3. Detects edges, textures, and shapes

4. Used in real applications

Recurrent Neural Network (RNN)

Recurrent Neural Network (RNN) is a type of neural network that is designed to work with sequential data, where the order of information matters.

Designed for Sequence Data

RNN works with data that comes one after another in a sequence.

Used for Text, Speech, and Time Series

Because it understands sequences, RNN is used in many applications such as: Text prediction, Speech recognition, Time series prediction

Has Memory of Previous Input

RNN has a memory feature. It remembers the previous input and uses that information to understand the next input.

Struggles with Long Sequences

RNN can remember recent information, but it often forgets information that appeared much earlier in the sequence.

Think of RNN like reading a story word by word while remembering the previous words to understand the sentence better.

Long Short-Term Memory (LSTM)

Long Short-Term Memory (LSTM) is a type of neural network used when data comes in a sequence, such as text, speech, or time-based data.

Special Type of RNN

LSTM is a type of RNN that processes data step-by-step and remembers important information for a long time.

Remembers Long-Term Dependencies

LSTM can keep important past information and use it later while making prediction

Used in Language Translation

LSTM helps translate languages by understanding the order and meaning of words in a sentence.

Think of LSTM like a smart notebook that remembers important information and forgets unnecessary details, helping AI understand sequences better.

Generative Adversarial Networks (GANs)

Generative Adversarial Networks (GANs) are a type of AI model where two neural networks work against each other to create new data.

Two Networks:

Generator & Discriminator

Tries to create fake data (like images).

Generator

Tries to detect whether the data is real or fake.

Discriminator

Real-World Uses

AI Art Generation

Deepfakes

Game Design

Medical Imaging

A teacher (Discriminator) checking if the painting is original or fake.

Imagine a student (Generator) drawing fake paintings

Daily Life Applications

Face Unlock → CNN

Stock Market Prediction → LSTM

Email Spam Detection → FNN

Google Translate → LSTM / RNN

AI Generated Images → GAN

Summary

Which neural network is best suited for generating new realistic images?

Which neural network is best suited for generating new realistic images?