Year · 2024–presentBy Ayush Niroula
A comprehensive deep learning framework built entirely from the ground up in Go. It features manual implementations of core neural network components including dense layers, ReLU/Softmax activation functions, and categorical cross-entropy loss, all powered by the Gonum numerical library for efficient matrix computations and data visualization.
Highlights
- Full implementation of neural network layers and activation functions without external ML frameworks.
- Custom loss calculation and accuracy metrics for multi-class classification tasks.
- Integrated dataset generators and visualization tools for benchmarking on complex patterns like spiral data.
- Optimized for performance through the use of Gonum's advanced matrix and numerical routines.
Tech stack
Go- Gonum
- Gonum Plot