A project for exploring perceptron neural network in Rust
The network aims to identify 'handwritten' digits from 0-9
(hopefully accurately) using the MNIST dataset
The library includes a drawing app designed to predict hand-written digits using the MNIST dataset. The current model achieves a 95% accuracy on both the training and testing MNIST dataset
- get matrix library up and running
- MNIST data loader
- get basic network structure
- drawing app
- model serialization with Serde
- momentum optimizer
- improve model accuracy / generalization
- rework matrix library trait abstractions
- create generic Data trait for feeding network data
This project is meant as a hobby. All the logic, including the matrix library, is implemented from scratch.