Our deep learning models run on-device. We have built models that run 10,000x faster than VGG’16 for only a <1% reduction in accuracy!