Researchers have introduced folded separable convolution architectures that restrict hardware requirements to two multipliers ...