AutoML for Model Compression and Acceleration on Mobile Devices using Reinforcement Learning

Abstract

Background: Model compression has been described as a crucial skill which resourcefully implement neutral network model on mobile devices possessing scarce computation assets and also operating under a tight budget. Most of the ancient model compression depend on methods which are handmade and also they operate under a rule-based procedure which only function under a domain expert so as to investigate one of the greatest design location for trading off for all the model size, speed, and the accuracy i.e. a sub-optimal and time consuming.