annikabrundyn/autodiff

Building a deep learning library from scratch

This project was an exercise to better understand the mechanisms like automatic differentiation that power deep learning libraries such as PyTorch, TensorFlow and JAX.

This implementation relies only on NumPy.

How to run

First, install the project:

cd autodiff
pip install -e .

Linked Colab Notebooks with demonstrations:

Comparing with PyTorch

Accelerating the convolution operator