[P] [D] Is inference time the important performance metric for ML Models on edge/mobile?

I am currently engaged on a project that aims to give some insight to machine learning engineers about how their models perform on vast variety of mobile devices.

It is starting to be a very popular practice to embed machine learning models within apps and use them without needing any api/network connection. You can see most examples especially for apps that use computer vision heavily. Passing each and every image to cloud for processing is simply unacceptable, data heavy and …

apps devices edge embed engineers inference insight machine machine learning machinelearning machine learning models ml models mobile mobile devices performance popular practice project vast