How do CNNs handle translation invariance and why is it important?

CNNs achieve translation invariance through convolution and pooling, allowing detection of patterns regardless of position.

Translation invariance means the model can recognize an object regardless of where it appears in the image. Convolutions detect features locally across the entire image, while pooling reduces sensitivity to exact positions. However, CNNs are not perfectly invariant; they are more accurately described as translation-equivariant with partial invariance introduced by pooling layers.