Overview

Image and grid data contain local structure that dense networks discard. This topic introduces convolutional networks as the parameter-sharing architecture that captures that structure efficiently, with applications extending from images to market-data grids and local temporal filters.

By the end of this note you should be able to:

• explain local receptive fields, shared filters, stride, padding, and pooling;
• compare the parameterisation of convolutional and dense layers;
• build a small CNN for grid-structured data.

1. Contents
2. The big picture