Democratizing machine learning will transform IoT
Even a few years ago, running machine learning on the battery-powered, wireless devices typical of IoT was extremely difficult for an embedded engineer to do. One of the main problems was machine learning required a PhD computer and data science background.
Today these restrictions no longer exist thanks to advances in both wireless IoT chips and a scaled-down version of machine learning called tiny machine learning or TinyML for short.
Although wireless IoT devices were typically described as resource constrained in the past, the same label cannot really be applied today at the leading edge of the industry. For example, in the Bluetooth industry an embedded dual-core processor can run at up to 128MHz supported by large amounts of flash memory and RAM. RAM is particularly useful for TinyML applications.
In machine learning, optimized neural networks can be shrunk down to run on the small computing resources. Reducing a high-level, computationally intensive AI or machine learning training model down into a TinyML version makes it possible for organizations to apply it to real-world IoT sensor-based applications. IoT applications focus on three key sensing areas: pattern and anomaly recognition in voice and audio, vibration, and vision in the form of digital images and video.
Put simply: Machine learning has become accessible to all in wireless IoT because of abstracting the machine learning complexity away from the end user, such as through graphical data representations.
Why is this such a big deal?
With TinyML, manufacturers can introduce local intelligence and decision making to almost any product or application. Doing any kind of AI or machine learning before this technology meant sending vast amounts of data up to the cloud for analysis. The power required for data analysis in the cloud tended to rule out battery operation, and the cloud was also very costly because it involved powerful web servers.
Machine learning at the edge, however, only sends essential data, such as an alert, up to the cloud. Even for continuous monitoring applications, the application can be in deep sleep mode most of the time to save battery and extend battery operation.
Countless products and applications can be made 10 to 100 times smarter and more useful in a way that hasn’t been considered technologically or commercially feasible before.
Doing as much data processing as possible at the edge also means greater levels of security and privacy can be achieved
The smart home
The smart home offers an example of an application that benefits from TinyML.
The smart home is a mess. Too many competing systems were incompatible with each other and smart home devices were just too difficult to set-up. Almost all of devices also demanded that you download an app, which wasn’t justified for the functionality many devices offered. Apps just cluttered the user’s smartphone.
The smart home industry has recognized this problem in the form of the Connected Home over IP (CHIP) initiative, an open standard that will allow full compatibility between various smart home products.
However, the use machine learning is how we will finally see the smart home fulfilling on its potential. Machine learning will make setting up and using smart home devices not only simple and indispensable in their usefulness.
Smart locks will move beyond app-based manual control to almost automatic self-control based on a household’s actual living patterns. If you forget to lock your front door when you usually lock it, the lock will lock itself. It will unlock at the times you take your dog for a walk each day. It may even give delivery people temporary access to leave parcels while you are out by opening an external door but locking an internal one to prevent access to the home.
Most importantly, the smart home will have to be smart on its own without sending data to the cloud for decision making.
Machine learning for the masses
With machine learning, IoT will become an indispensable part of everyday lives, both private and professional.
End users won’t necessarily care about the machine learning details, but they will care about the results in terms of effortless, intuitive interactivity, seamlessly intelligence and useful operation. Many set-up and maintenance tasks will be eliminated, too.
Thousands of TinyML-powered applications and products will emerge over the coming years. And these will all work together to make the world a bit safer, more comfortable, more efficient, less prone to breakdown and delays, more economical in its consumption of scarce resources and less wasteful.
All IoT Agenda network contributors are responsible for the content and accuracy of their posts. Opinions are of the writers and do not necessarily convey the thoughts of IoT Agenda.