Machine Learning with Swift

MobileNets

MobileNets is a class of efficient CNNs targeted to mobile and embedded applications. It was proposed by the Google research team in MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications, 2017. In comparison to the traditional CNN, it has much less parameters, and requires much less computation of the learning and prediction process. This makes it faster and lighter, preserving the predictions accuracy at the same time. The main innovation was the introduction of depth-wise separable convolutions: http://machinethink.net/blog/googles-mobile-net-architecture-on-iphone/.

The original paper can be found here: https://arxiv.org/abs/1704.04861.