Why do we actually need convolutions? Can’t we use fully-connected layers for that?
Computer VisionLevel: IntermediateNeural networks for computer vision
Why do we actually need convolutions? Can’t we use fully-connected layers for that?
Hints
- Track the receptive field and input/output dimensions.
- Check stride, padding, weight sharing and channel count.
Answer
A fully-connected layer needs one weight per inter-layer connection, which means the number of weights which needs to be computed quickly balloons as the number of layers and nodes per layer is increased.
Source and licence: Alexey Grigorev & contributors — Data Science Interviews · CC-BY-4.0