alexanderwangy / mnist_from_scratch

A deep learning handwritten digit classifier made using PyTorch, TF Keras Datasets. All basic calculations and transformations are done without a pretrained model or a framework made model.

MNIST Model from scratch (with help from PyTorch)

This is a hand drawn digit classifier deep learning "model". It was made with help from PyTorch, especially their tensors and matrix multiplications (although in truth it was just the '@' that I used). You can check out the notebook linked on this GitHub repository.

Some future adjustments:

• Add a convultional mask before training to add more attention to corners and edges
• Possible blur? How does that affect accuracy
• Add benchmarks for different combinations of CNs and Blurs and Dense Layers
• Add more Dense Layers?
• Make it customizable, so you can remove and add layers as you wish.