Why is a CNN better for image-based tasks?
Name three reasons why a convolutional neural network is better for image-based tasks than a basic MLP.
Answer
Three main reasons why CNNs are better for image-based tasks than basic MLPs:
CNNs preserve spatial structure: Unlike an MLP, which usually flattens an image into a long vector, a CNN keeps the image as a 3D tensor.
This lets it understand local (pixel) relationships, such as edges, corners, shapes, etc.
CNNs use local filters: This is useful because visual patterns are usually local, such as edges, corners, shapes, etc.
CNNs are robust to translations: A CNN can detect a feature/object/shape even if it appears in different places in the image.