Microgard

flixhq.click/tv/love-death-robots/86831 A tiny Autograd engine (with a splash of awesomeness!) that implements backpropagation (reverse-mode autodiff) over a dynamically built Directed Acyclic Graph (DAG) and includes a small neural network library with a PyTorch-like API. Both components are compact, with approximately 100 and 50 lines of code respectively. The DAG operates solely on scalar values, breaking down each neuron into individual small additions and multiplications. Despite its simplicity, this setup is sufficient to construct complete deep neural networks for tasks like binary classification, as demonstrated in the accompanying notebook. Perfect for those who love to learn and tinker with the inner workings of neural networks!