In mathematics, a convolution for 1D (e.g. time series) or 2D (e.g. image) data is a set of linear weights that are applied to the neigbours of each point to obtain a new series or set of 2D values. For example, given a time series ti, we can create a new series si where:
si = 0.25 * ti + 0.5 * ti-1 + 0.25 * ti-2
This concept was the inspiration for convolutional neural networks, where instead of a set of linear weights (such as (0.25,0.5,0.25)) the same small neural network is trained and used.
Used on page 272