Enhanced Capability on Smart Handheld Devices Using On-Device Machine Learning

The current, most popular method for making edge devices intelligent relies on the cloud, where data is gathered from numerous device sources and uploaded. The developed model is then sent back to the device after being used to train a machine learning model in the cloud. The device then uses this trained model & becomes even more intelligent. Given recent developments and a growth in the number of smart devices with improved hardware capabilities, there is an increasing interest in using machine learning on the edge device itself rather than learning in the cloud. Hardware vendors are promoting AI-enabled chipsets that offer improved processing capabilities better tailored to computer vision, IoT, and machine learning based applications and solutions that benefit end users, which is further fostering this interest. This idea also reduces the overhead of offloading data to the cloud every time & also solves the security concern of user data being offloaded to the cloud. This way an enhanced security will be assured for user data and reduce overall latency for certain time-critical machine learning applications. This research study has surveyed the benefits and challenges of implementing Machine learning algorithms on edge devices by paying special attention to how techniques are adapted or designed to execute the resource-constrained devices.

I. Introduction

With latest advancement in smartphone processors, chipset vendors had introduced on device capability for machine intelligence & cognitive capabilities. With this, the OEM will give mobile applications cognitive features including voice, audio, and visual perception. This is still in beginning stage & not much is adopted in day-to-day life due to accuracy of data and processing of that data.