Learning archive · Educational project
Handwritten Digit Recognition
A small CNN, from training to inference.
An educational convolutional classifier for MNIST digits with saved-model inference for custom images.
The question
A classifier must connect a reproducible training/evaluation split to the preprocessing used at inference.
How it works
- Train a convolutional model on the MNIST training split.
- Evaluate on the separate test split.
- Save the model and prepare custom digit images for inference.
Evidence & results
- Stored project result: 99.1% accuracy on 10,000 MNIST test images after training on 60,000 images.
Scope & next steps
- Historical educational dataset result; not a new benchmark run or evidence of unconstrained handwriting recognition.
Go one level deeper.
Ask about the implementation, tradeoffs, or my contribution.
Explore with the portfolio guide