Micrograd: A Tiny Autograd Engine
Developed a small automatic differentiation engine that implements backpropagation over a dynamically constructed directed acyclic graph.
A lightweight neural network library was built on top of the engine with an API inspired by PyTorch.
Key Features:
- Reverse-mode automatic differentiation
- Dynamically constructed computation graph
- Backpropagation
- Basic neural network layers
- PyTorch-inspired API
Technologies: Python, Neural Networks, Automatic Differentiation