🕸️ Neural Networks for CIFAR-10 Classification

This project involves implementing both a Fully Connected Neural Network (FCNN) and a Convolutional Neural Network (CNN) for classifying images from the CIFAR-10 dataset using TensorFlow. The main goal is to train and test these networks, while also performing hyperparameter search to optimize model performance.

Table of Contents

• Setup
• Problem Statement
• Fully Connected Neural Network
• Convolutional Neural Network
• File Structure
• Submission

Setup

Prerequisites

This project uses Python for implementation. Ensure that you are working within a virtual environment to avoid dependency issues.

1. Create a new virtual environment:

   python -m venv env

2. Activate the virtual environment:

   source env/bin/activate

3. Install the required dependencies:

   pip install tensorflow

Dataset

We use the CIFAR-10 dataset for this project, which contains 50,000 32x32 color images across 10 classes. You can find more information about the dataset here: CIFAR-10 Dataset.

Problem Statement

In this project, I have implemented a Fully Connected Neural Network (FCNN) and a Convolutional Neural Network (CNN) to classify images from the CIFAR-100 dataset.

1. Fully Connected Neural Network:

• Implement a neural network with three linear layers.
• Complete the forward pass and apply the activation function (e.g., ReLU) to the first two layers.
• Train and test the model, then report the test accuracy.

2. Convolutional Neural Network:

• Implement a CNN with three convolutional layers, followed by max pooling and a flatten layer.
• Apply activation functions to the convolutional layers.
• Complete the forward pass and ensure the model is correctly trained and tested.

3. Hyperparameter Search:

• Explore various hyperparameters to optimize model performance.