Lecture
This lecture covers the basic ideas of automatic differentiation, focusing on the concept of reverse mode differentiation to optimize the filters of a convolutional layer by gradient descent. The process involves recording the computation tree on the forward path to apply the chain rule in reverse order, similar to backpropagation in a simple feedforward network. Examples of automatic differentiation are provided, showcasing the determination of children nodes of weight variables, backward scheduling, and the use of primitive operations.