SOFTWARE & SYSTEMS
MNIST FROM SCRATCH
A neural network for MNIST handwritten-digit recognition implemented from scratch using only NumPy - no ML frameworks.
completed
- Year
- 2019
- Language
- Python
Handwritten-digit recognition · Concept illustration
NEURAL NETWORK USING ONLY NUMPY
01 / The build
Behind the project.
A neural network that recognizes MNIST handwritten digits, implemented from scratch in a single script (mnist.py) using only NumPy: forward pass, backpropagation and training are all written by hand, without any machine-learning framework.
Technical details
Under the hood.
- Language
- Python
- Dependencies
- NumPy only
- Dataset
- MNIST handwritten digits
Notes & outcomes
What came out of it.
- Forward pass and backpropagation written by hand.
- No TensorFlow, PyTorch or any ML framework - NumPy only.
- Single-file implementation (mnist.py).