Hands-On Transfer Learning with Python

The convolution operator

The central idea behind CNNs is the mathematical operation of convolution, which is a specialized kind of linear operation. It's heavily used in a variety of domains, including physics, statistics, computer vision, and image and signal processing. To understand this, let's start with an example. A noisy laser sensor is tracking the position of a spaceship. To get a better estimate of the location of the spaceship, we can take an average of a few readings, giving more weights to recent observations. Let x(t) represent the location in time, t, and let w(t) be the weighting function.

The estimate of the position can be written as:

Here, the weight function, w(t), is called the kernel of the convolution. We can compute ...