AI Blog

AI Blog

by Michele Laurelli

Convolution

/ˌkɒnvəˈluːʃən/
Concept
Definition

A mathematical operation that slides a filter/kernel over input data to extract features.

Convolution computes dot products between the filter and local regions of input. The filter moves with a stride, producing a feature map. Key operation in CNNs for spatial feature extraction.

Examples

1

2D convolution for images

2

1D convolution for sequences

3

Edge detection filters

Michele Laurelli - AI Research & Engineering