A toy example and convolutional layers (part 1)

To introduce some of the basic components in a CNN, we make use of a video by Brandon Rohrer Links to an external site. (we have separated his video into "A toy example and convolutional layers" and "Components of CNNs" and introduced quizzes). The video makes use of a toy example to illustrate how convolutions can be used to extract features and make decisions. Toy examples are great to build intuition!

In the first half (A toy example and convolutional layers), the main learning objective is: 

• How different filter kernels can extract different features. 
   

One misleading detail in this part of Brandon's video: 

• When computing convolutions he divides the values by the size of the filter kernel. That is, for a 3x3 kernel he divides the output by 9. This is not standard and only done here for illustration purposes.