There were some interesting attempts to get smaller models running on device post SqueezeNet. What was needed was a model designed specifically on mobile devices. What a group of researchers at Google produced was called MobileNet, which is an important family of networks for you to understand and where we will be spending a few chapters. At a high level, we will use depthwise separable convolutions to produce an even more accurate network than SqueezeNet that runs well on mobile phone hardware.
MobileNet (v1)
> MobileNets: Efficient Convolutional Neural Networks for Mobile ...
