/mnist-model-parameters2

Trained weights and biases for a PyTorch model on the MNIST dataset, aimed at handwritten digit recognition

Primary LanguageJavaScriptMIT LicenseMIT

MNIST Model Parameters

This repository provides pre-extracted weights and biases from a model trained on the MNIST dataset, specifically designed for handwritten digit recognition. The parameters are formatted in JavaScript (JS) to facilitate seamless integration into web applications or any JavaScript-based project.

Overview

The MNIST ("Modified National Institute of Standards and Technology") dataset is a cornerstone in the field of machine learning, providing a large set of handwritten digits for training image processing systems. This repository leverages a Convolutional Neural Network (CNN), trained on the MNIST dataset using PyTorch, to offer a ready-to-use set of model parameters.

Features

  • Pre-Extracted Parameters: Includes weights and biases for three layers of the CNN, stored in separate JS files for ease of use.
  • JavaScript Compatibility: Parameters are provided in a JavaScript-friendly format, enabling straightforward integration with web technologies.
  • High Performance: The model architecture and training have been optimized to achieve high accuracy on digit recognition tasks.

Repository Structure

  • /weights: Contains JS files (weightsLayer1.js, weightsLayer2.js, weightsLayer3.js) with the model's layer-specific weights.
  • /biases: Includes JS files (biasesLayer1.js, biasesLayer2.js, biasesLayer3.js) with the biases for each corresponding model layer.
  • LICENSE: The MIT License file detailing the terms under which the repository's contents can be used.

Getting Started

To integrate the MNIST model parameters into your project, simply include the relevant JS files from the /weights and /biases directories in your application. Detailed instructions and examples of how to use these parameters within your projects are forthcoming in future updates to this repository.

Performance Metrics

The model achieves an accuracy of 98.7% on the MNIST test set, making it highly reliable for digit recognition tasks. Additional metrics such as precision, recall, and F1 score are also indicative of the model's robust performance.

License

This project is released under the MIT License, which provides extensive freedom for reuse within your own projects, subject to the original copyright notice and license agreement. See the LICENSE file for full details.

Stay tuned for further updates and enhancements to this repository!