Hi there,
I have made this short 5 min video summarizing about clever optimizations that yield massive improvements on Image recognition training, and one is Mixed Precision usage:
Video: Supercharging AI with high performance distributed computing - YouTube
Paper: [1811.06992] Image Classification at Supercomputer Scale
Will you please see it and give it a “thumb up” if you like it.
Comments welcome (specially feedback) and don’t forget to share.
Hope it’s helpful / Thanks and regards.