Deep Learning with PyTorch Quick Start Guide

A single convolutional layer

Convolutional layers are organized so the units in the first layer only respond to their respective receptive fields. Each unit in the next layer is connected only to a small region of the first layer, and each unit in the second hidden layer is connected to a limited region in the third layer, and so on. In this way, the network can be trained to assemble higher level features from the low-level features present in the previous layer.

In practice, this works by using a filter, or convolution kernel, to scan an image to generate what is known as a feature map. The kernel is just a matrix that is the size of the receptive field. We can think of this as a camera scanning an image in discrete strides. We calculate ...