May The Course Be With You! ML101 Episode3b | Backpropagation in Neural Networks скачать в хорошем качестве

May The Course Be With You! ML101 Episode3b | Backpropagation in Neural Networks

May The Course Be With You! I’m Mark Tschopp, and in the second part of this Neural Network Basics series, I explain backpropagation---the backbone of how the model parameters for these neural networks are trained. So, what is backpropagation? It really boils down to how you compute the derivative term within gradient descent. Remember the chain rule of calculus? This allows the gradient term to start at the output node and work its way back to model parameters throughout the neural network. The derivative of the linear part, of the activation function, all come together in both a mathematics-oriented example as well as a more visual representation of the concept.