CNN: Powering Visual Intelligence in AI

This sesssion introduced the fundamentals of Convolutional Neural Networks (CNNs) and demonstrated how deep learning models can perform image classification tasks using TensorFlow and Keras.

📂 Contents

1️⃣ MNIST_completed.ipynb – MNIST Digit Classification

• Dataset: MNIST
• Built a CNN to classify handwritten digits (0–9).
• Covered preprocessing (reshaping, normalization), convolution layers, pooling, and fully connected layers.
• Trained and evaluated the model, achieving high accuracy on the test set.
• Included visualization of predictions and pixel grids for conceptual clarity.

2️⃣ Cat_Dog_FIXME.ipynb – Binary Image Classification

• Custom train/test dataset (Cats vs Dogs).
• Implemented CNN with data augmentation using ImageDataGenerator.
• Applied convolution, pooling, flattening, dense layers, and sigmoid output.
• Trained the model and performed single-image predictions.

📊 Slides – Pixels to Patterns

The presentation covered:

• Why traditional ANNs struggle with image data
• How CNNs mimic human vision
• Convolution, filters, feature maps, and pooling
• End-to-end CNN pipeline

Slides focused on building intuition before implementation.