Deep Learning Systems: Neural Networks and Training at Scale

<p>Follow Mara Okafor, a newly hired ML engineer at Verdant Robotics, a fictional startup building AI-guided recycling sorting arms, as she replaces a failing rules-based vision system with a real neural network. Starting from a single neuron built by hand in numpy, Mara works up through backpropagation and PyTorch's autograd, builds a convolutional image classifier for sorting recyclables, and debugs the exact problems every deep learning engineer eventually hits: a loss that won't move, vanishing gradients in a deep sigmoid network, an overfitting model that fails on the production line, and the scaling changes &mdash; batching, mixed precision, and multi-GPU training &mdash; needed as the system grows from one facility to three.</p>

Tools: PyTorch, Neural Network Fundamentals, Backpropagation, Optimizers, Regularization, Training at Scale