Neural Network Gradients: Backpropagation, Dual Numbers, Finite Differences (blog.demofox.org)