https://github.com/mohammedsaqibms/initialization
This repository explores the impact of various weight initialization methods on a neural network's performance, comparing zero, random, and He initialization. It includes visualizations of cost function and decision boundaries.
https://github.com/mohammedsaqibms/initialization
ai deep-learning he-initialization machine-learning neural-networks weight-initialization
Last synced: about 1 year ago
JSON representation
This repository explores the impact of various weight initialization methods on a neural network's performance, comparing zero, random, and He initialization. It includes visualizations of cost function and decision boundaries.
- Host: GitHub
- URL: https://github.com/mohammedsaqibms/initialization
- Owner: MohammedSaqibMS
- Created: 2024-09-11T13:56:30.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2025-02-19T15:48:55.000Z (over 1 year ago)
- Last Synced: 2025-02-19T16:39:23.659Z (over 1 year ago)
- Topics: ai, deep-learning, he-initialization, machine-learning, neural-networks, weight-initialization
- Language: Jupyter Notebook
- Homepage:
- Size: 294 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Neural Network Initialization and Training 🧠🚀
Welcome to the Neural Network Initialization and Training project! This project demonstrates the importance of initialization methods when training deep learning models. It showcases how different initialization strategies affect the model's learning and performance. We'll cover three key initialization methods:
- **Zero Initialization** ➡️ All weights set to zero.
- **Random Initialization** ➡️ Weights are randomly initialized.
- **He Initialization** ➡️ Weights are initialized using He et al.'s method, suitable for ReLU activations.
## 🙏 Credits
This project is based on concepts and methodologies taught in the [Deep Learning Specialization](https://www.deeplearning.ai/courses/deep-learning-specialization/) by **Andrew Ng** and **DeepLearning.AI**.