Generally, IoT systems can be divided into two groups: those applied merely for monitoring tasks (tend to be very data-intensive), and those meant to enable remote control of “intelligent” devices.
The first ones are usually analytics solutions that aggregate information from a multitude of IoT sensors (data comes up through gateways) and then format or visualize it to help users discern patterns and actionable insights from their datasets. They may have vast bandwidth requirements when there are plenty of devices incorporated, but only in terms of upstream bandwidth.
The latter, however, have much more difficult data models and both their upstream and downstream bandwidth are equally critical; there’s always a complex business logic involved in such platforms that’s used to 1) determine the needed adjustments based on received data and 2) send updated configurations to the devices to optimize performance.
The IoT networks of the second type, popular with enterprises, are typically comprised of a great many entities and scaling them safely, by introducing new devices and data streams, is acutely challenging. Vulnerabilities accumulate exponentially as new elements and relationships are added to the network.
In this post, we’ll discuss how to enhance a conventional IoT architecture through machine learning. We’ll explain how supervised, unsupervised, deep and reinforcement learning methods could help firms protect their connected solutions against various types spoofing, jamming, intrusion, man-in-the-middle, dos and other identity-based attacks.