Decentralized AI and IoT Edge devices
Centralized AI:
Over the past years, tech giants have been working on artificial intelligence and Internet of Things (IoT) leveraging the computational power of cloud computers. Google, Microsoft, Amazon, IBM and many more, all are investing in cloud based AI, IoT and machine learning. So far this concept of centralized AI has been working well and bringing in marginal return on investment (ROI). Traditionally the cloud and AI based IoT systems that we familiar with are,
- Distributed sensor based data collection
- Mobile asset monitoring and tracking
- Controlling assets through web APIs and services
However, in these kind of IoT architectural designs cloud is taking most of the data analysis and computational loads but micro-controllers or sensors only collecting the raw data and sending the raw data to the cloud through web APIs, mainly using REST (Representational State Transfer) based APIs. Siting in the center, the cloud receives the data, performs the stream analytics on the raw data, runs the business logic and machine learning, and finally takes an intelligent decision. This decision goes back to the device as instructions. Cloud is doing all the computational work and carrying out intelligent response in centralized AI based IoT systems. Sounds good and fits perfectly… but there are some issues!
MCU or micro-controller units are re-programmable, such as Raspberry PI, Arduino etc. These MCUs are only sending data to cloud through APIs. But the communication latency is high, sending data through APIs for a single sensor read increases the latency hence downgrades the quick response of an AI system. MCU needs to wait for the decision taken by centralized AI. All the data analysis and decision making are done in cloud.
Decentralized AI:
Decentralization of AI bringing a new trend in IoT systems by giving the responsibility of primary computing in edge devices. By term, IoT edge devices are capable of performing stream analytics of real-time data, machine learning and also take decision based on learning outcomes. This decentralizes the AI systems and distributes the intelligence to edge devices.
This is bringing a remarkable progress in automation. Microchips and hardware manufacturing companies are also foreseeing a profitable market in IoT edge computing. Intel, Qualcom, AMAT etc. are designing computers and microchips for IoT edge computing. Alongside the hardware, Google and Microsoft are building IoT edge SDKs (Software Development Kit). Developers would be able to use the SDKs to transform the devices into individual AI units and allow devices to take collective decisions in decentralized AI network. This design is letting the edge IoT devices to use battery power in computation instead of device to cloud communication. Its increasing the privacy of the communication, reliability, latency and making the system power-efficient.
Decentralization of AI and computing on edge for IoT devices would require edge devices with more AL/ML capability, application processors and controller section in the same core package, dedicated real-time processing and more. But clearly in decentralized AI, we can visualize a mesh network of IoT edge devices taking decisions as a whole other than fully depending on a centralized cloud for any decision.
Director at Logical Line Marking
6 年I’ve always been impartial to AI, but you’ve got me thinking now…