Slide 1: Title Slide

(Pause briefly to introduce yourself if needed.)

Hello everyone . Today, I will be presenting on Convolutional Neural Networks (CNNs), Recurrent Neural
Networks (RNNs), and Long Short-Term Memory (LSTMs).

These models are fundamental in deep learning and are widely used in image recognition, speech
processing, and time-series forecasting. Let’s begin!

Slide 2: Convolutional Neural Network (CNN)

What is CNN?

A Convolutional Neural Network (CNN) is a deep learning model specifically designed for working with
images and videos.

Unlike traditional neural networks, CNNs automatically detect patterns and features in an image,
making them ideal for tasks like image classification, object detection, and video analysis.

Slide 3: CNN Components

CNNs have four main components:

1. Convolution – Extracts important patterns.

2. Non-Linearity (ReLU) – Introduces non-linearity to handle complex data.

3. Pooling (Subsampling) – Reduces dimensions but retains key information.

4. Fully Connected Layer – Classifies the final output.

Slide 4: How CNN Works

CNNs process an image through different layers to extract meaningful information. The key layers are:

 Input Layer

 The input is an image, represented as a grid of pixel values.

 If it's grayscale, each pixel has a value between 0 and 255.
 If it's RGB, each pixel has three values (Red, Green, Blue).

 Convolution Layer

 This layer detects features like edges, textures, and shapes using small filters (kernels).
 These filters slide over the image, creating a feature map that highlights important
patterns.
 Activation Function (ReLU - Rectified Linear Unit)

 ReLU removes negative values, making the model more efficient.

 It helps CNNs detect complex patterns in images.

 Pooling Layer (Downsampling)

 Reduces the size of the image to make processing faster.

 Max Pooling keeps only the most important values from each region.
 Average Pooling takes the average of nearby values.

 Fully Connected Layer

 The extracted features are passed to a fully connected layer (like a traditional neural
network).
 This layer makes the final classification of the image.

Slide 10: Different CNN Models

Some well-known CNN architectures are:

LeNet: A pioneering CNN architecture by Yann LeCun, designed for handwritten digit recognition
(MNIST).

AlexNet: A deep CNN that won ImageNet 2012, introducing ReLU activation and dropout for better
training.

ResNet: A deep network using residual connections (skip connections) to tackle vanishing gradients in
training.

GoogleNet: A CNN with an Inception module, optimizing computational efficiency and accuracy.

VGG: A deep CNN with uniform 3x3 convolutional layers, known for simplicity and strong feature
extractionEach model improves accuracy and efficiency in image recognition.

Slide 11: Applications of CNN

CNNs are widely used for:

✅ Image classification (Face recognition, self-driving cars)
✅ Object detection (Detecting humans, animals, or objects in images)
✅ Image segmentation (Medical image analysis)
✅ Video analysis (Motion tracking, security cameras)
Slide 12: Advantages of CNN

CNNs offer:
✅ High accuracy in image-related tasks.
✅ Efficiency by reducing manual feature extraction.
✅ Robustness in detecting objects in different conditions.

Slide 13: Introduction to Recurrent Neural Networks (RNNs)

A Recurrent Neural Network (RNN) is a deep learning model designed to process sequential data, such
as text, speech, and time-series information.

Unlike CNNs, which work on images, RNNs remember past information while processing new inputs.

Slide 14: RNN Architecture

 RNNs have a hidden state that stores past information.

 They process data step by step, remembering previous inputs.

Slide 15: How RNN Works

1. Takes an input (e.g., a word in a sentence).

2. Combines it with past memory to understand the context.

3. Updates its hidden state and predicts the next output.

This makes RNNs great for tasks like speech recognition and text prediction.

Slide 16: Variants of RNN

Different types of RNNs include:

✅ Vanilla RNN – Basic model but suffers from memory loss.
✅ Bidirectional RNN – Processes data in both forward and backward directions.
✅ LSTM & GRU – Solve memory problems in standard RNNs.

Slide 17: Applications of RNN

RNNs are used in:

✅ Time-Series Forecasting (Stock market prediction, weather forecasting)
✅ Natural Language Processing (NLP) (Text generation, chatbots)
✅ Speech Recognition (Voice assistants like Siri and Alexa)
Slide 18: Introduction to LSTM

Long Short-Term Memory (LSTM) is a special type of RNN designed to handle long sequences of data
without losing important information.

LSTMs solve the memory loss problem in standard RNNs.

Slide 19: LSTM Architecture

LSTMs have four key components:

1. Cell State –It is the first component of LSTM which runs through the entire LSTM unit. It kind of
can be thought of as a conveyer belt.

This cell state is responsible for remembering and forgetting. This is based on the context of the
input. This means that some of the previous information should be remembered while some of
them should be forgotten and some of the new information should be added to the memory.
The first operation (X) is the pointwise operation which is nothing but multiplying the cell state
by an array of [-1, 0, 1]. The information multiplied by 0 will be forgotten by the LSTM. Another
operation is (+) which is responsible to add some new information to the state.

2. Forget Gate – Decides what to forget.

It does a dot product of h(t-1) and x(t) and with the help of the sigmoid layer, outputs a number
between 0 and 1 for each number in the cell state C(t-1). If the output is a ‘1’, it means we will
keep it. A ‘0’ means to forget it completely.

3. Input Gate – Determines what new information to store.

4. Output Gate – Selects what to send as output.

Slide 21: Applications of LSTM

LSTM is used in:

✅ Language Modeling (Text generation, chatbots)
✅ Speech Recognition (Translating speech to text)
✅ Recommender Systems (Personalized recommendations on YouTube, Netflix)

Slide 22: CNN vs. RNN vs. LSTM

Feature CNN RNN LSTM

Best for Images & objects Sequential data Long-term sequences

Feature CNN RNN LSTM

Memory No memory Short-term memory Long-term memory

Use Case Image classification Chatbots, speech-to-text Time-series forecasting, NLP

Slide 23: Conclusion

✅ CNN is best for images and object detection.

✅ RNN is best for sequential data like speech and text.
✅ LSTM improves RNNs by handling long-term memory.

These models power AI applications like Google Translate, self-driving cars, and virtual assistants.

Slide 24: Thank You!

Thank you for your time! I hope you found this presentation informative. Do you have any questions?