7 Deep Learning: tensor backpropagation, convolutional layers (MLVU2020)

slides: course materials: We recap the basic algorithm of (scalar) backpropagation from last lecture and we discuss in detail what is required to extend this into a mature automatic differentiation system: how to deal with any computation graph, and how to apply backpropagation to models that operate in terms of tensor operations (like matrix multiplication). The explanation broadly follows the structure of PyTorch, but