NNFS Extreme
NNFS Extreme
A C++ neural network engine. Training and inference. 98.34% accuracy on MNIST. No external dependencies.
Purpose
Build a neural network from the mathematics up, implementing the matrix operations, training loop, and inference engine directly.
Implementation
- Custom linear algebra, mini-batch training, and digit classification in C++.
- Python bindings connect the engine to an OpenCV and MediaPipe air-drawing experiment.
- Grid search harness which tested 1152 different configurations with combinations of learning rates, mini-batch sizes, topologies, etc.
