ML//CNN//convolution

- A filter slides across the input, dot product at each position.

A filter slides across the input, dot product at each position.

Small filter (3×3, 5×5) learns local patterns — edges, corners, textures.

Shared weights across positions = translation invariance.