SensiML™ Corporation, a leading developer of AI tools for building intelligent IoT endpoints, has released a white paper entitled “The Opportunity for AI at the Edge and Beyond.” This comprehensive review explains how Internet of Things (IoT) endpoints have, until very recently, been focused on acquiring raw sensor data and feeding the associated data back towards the “core” of the network for processing (whether local or cloud connected). However, two significant trends and a host of related issues are beginning to drive AI and Machine Learning (ML) towards the edge of the network and in many cases onto the actual endpoint devices themselves. The first trend is that the number of IoT nodes is increasing dramatically each year. The second trend is that the amount of data being generated by each device is also increasing significantly.
Moving AI to the edge of the network is highly desirable in many cases as it often improves the following application characteristics:
- Autonomy: Independent insight and control is much easier if decisions are made locally
- Reliability: Dependency on cloud connections can be reduced or eliminated
- Security/Privacy: Risk of raw data interception significantly lowered
- Efficiency: Amount of data to be transmitted across the network is vastly reduced
- Responsiveness/Latency: No waiting while data is transmitted to cloud and back
Conversely, implementing intelligence at the extreme edge or IoT device has its own set of challenges. With processing and memory resources many of orders of magnitude smaller than that available in the data center, computational approaches popular in cloud AI are not well suited to most edge applications. Adapting cloud AI to run efficiently and fit on edge devices often involves hand-coding which is time and resource intensive and often impractical.
The Opportunity for AI at the Edge and Beyond whitepaper explains these trends and issues comprehensively, but in an easy-to-understand format. It goes on to discuss applications which can benefit from moving AI to the edge of a network, the challenges associated with practical implementations, and possible solutions of new AutoML tools available from various vendors.