https://github.com/asherk7/neural-network-from-scratch
Creating a Convolutional Neural Network from scratch, modeled after ResNet
https://github.com/asherk7/neural-network-from-scratch
classification cnn ml neural-network numpy python resnet
Last synced: 26 days ago
JSON representation
Creating a Convolutional Neural Network from scratch, modeled after ResNet
- Host: GitHub
- URL: https://github.com/asherk7/neural-network-from-scratch
- Owner: asherk7
- License: mit
- Created: 2025-04-29T22:57:30.000Z (about 1 year ago)
- Default Branch: main
- Last Pushed: 2025-05-20T00:30:03.000Z (about 1 year ago)
- Last Synced: 2025-08-18T15:39:02.630Z (10 months ago)
- Topics: classification, cnn, ml, neural-network, numpy, python, resnet
- Language: Python
- Homepage:
- Size: 36.1 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Convolutional Neural Network (CNN) from Scratch
Creating a Convolutional Neural Network (CNN) from scratch using Python and NumPy, model is a simplified version of the ResNet architecture from the paper ["Deep Residual Learning for Image Recognition"](https://arxiv.org/abs/1512.03385).
## Overview
This repository contains an implementation of a convolutional neural network and the popular components, built entirely from scratch using Python and NumPy, modeled after a simple ResNet. The goal of this project is to gain a deep understanding of the inner workings of neural networks, including:
- Forward propagation
- Backward propagation
- Activation functions
- Loss functions
- Optimization techniques
- Residual connections
The network will be trained and evaluated on the **CIFAR-10 dataset**, a widely used benchmark for image classification tasks.
---
## Features
- Manual implementation of key neural network components:
- Multiple layers
- Non-linear activation functions
- Loss functions
- Gradient descent and backpropagation
- Residual connections
- Training and evaluation pipeline for the CIFAR-10 dataset
---
## Objectives
1. Understand the mathematical foundations of neural networks.
2. Implement a neural network without relying on high-level libraries like TensorFlow or PyTorch.
3. Build intuition on how different components work together.
---
## Prerequisites
- **Python 3.8+**
- **NumPy 2.2.1**
- Basic knowledge of linear algebra, calculus (derivatives & multi-variable), and programming.
---
## Installation
1. Clone this repository:
```bash
git clone https://github.com/asherk7/neural-network-from-scratch.git
cd neural-network-from-scratch
```
---
## Usage
1. Make sure you have the required libraries above.
2. Download the CIFAR-10 dataset, rename it to `data`, and place it in the same directory as the script. The dataset can be downloaded from [here](https://www.cs.toronto.edu/~kriz/cifar.html).
3. Run the cifar script to train the model:
```bash
python cifar.py
```
4. The model will be trained on the CIFAR-10 dataset, and you can monitor the training process through the printed metrics.
5. After training, the model's performance will be evaluated on the test set, and results will be displayed.
---
## References
- [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385)
- [Build a Neural Network from scratch in C++ to deeply understand how it works, not just how to use it](https://medium.com/@sirawitchokphantavee/build-a-neural-network-from-scratch-in-c-to-deeply-understand-how-it-works-not-just-how-to-use-008426212f57)
- [Neural Networks from Scratch in Python](https://www.youtube.com/playlist?list=PLQVvvaa0QuDcjD5BAw2DxE6OF2tius3V3)
- [CIFAR-10 Dataset](https://www.cs.toronto.edu/~kriz/cifar.html)