330772 (1) [Avatar] Offline
In Section 5.1.1 it is mentioned that convolutions help the neural net learn translation-invariant patterns (i.e., a pattern that appears in one part of the image can be recognized in another part of the image). It may be worth noting that the activations in the convolutional layer are not translation-invariant but are translation-equivariant, as explained on Slide 4 in https://www.cs.toronto.edu/~hinton/csc2535/notes/lec6a.pdf.