TinyML is an embedded software technology that is made to run on very limited hardware. This device has got a tiny camera, and a visual application to take pictures and to perform object detection in a reasonable amount of time. Image: Telenor

How to make tiny sensors smart – Telenor Research is working for the next IoT revolution

TinyML is a hot topic on the research agenda at Telenor Research for many good reasons. It is a perfect match with NB-IoT, a mobile network optimized for IoT. It has infinite real-life use cases, for instance, in cars, agriculture, transport, and tourism. Do you know what TinyML stands for, and why it is considered revolutionary in IoT development?

Since 2019, DNA has been a part of Telenor Group, one of the largest operators in the Nordic countries and world's leading suppliers of IoT solutions. Telenor invests in strong Nordic co-operation and provides synergies and economies of scale for its Nordic organizations such as DNA. In practice this means sharing research findings and developing joint offering for the Nordic B2B customers. Through this collaboration, Telenor research services, research outcomes, and business insights are also available to DNA and its customers.

Artificial intelligence innovations are bread and butter for Vegard Edvardsen and Gard Spreemann, Research Scientists at the AI and analytics department of Telenor Research. In general, they work closely with customer services, customer relations, and network operations. Their team tries to find new use cases that can be relevant for Telenor Group.

“Our interest is to understand what is possible, what solutions would be beneficial to certain industries, and what creates new growth. We see IoT as one area for growth, and we need to study how to make IoT useful to us and our customers. That is why we are looking what AI means for the Internet of Things,” Vegard Edvardsen says.

It is a tough challenge to combine AI and IoT. Normally, artificial intelligence requires heavy computing resources, and its processes run in data centers. IoT (Internet of Things) represents the opposite end of the spectrum: devices are small, and their computing capability is extremely limited. Typically, they are placed in the “edge”, or field environments, to gather sensor data independently and require a very long battery life.

Solving this challenge explains why TinyML is an innovation on Edvardsen and Spreemann’s current research agenda – and it makes them really tick. What is it?

How tiny is TinyML?

“TinyML is what happens, when you take AI and IoT, and bring AI computing to the edge. It means tiny devices that run simplified deep learning models and still do some sophisticated work,” Gard Spreemann tells. “TinyML is a research breakthrough, which has grown from an academic curiosity to a whole new field of research with a lot of traction.”

The acronym ML stands for machine learning software. Tiny hints that it is made to run on very limited hardware. To illustrate the physical aspect, Vegard Edvardsen picks up something from a table. It is smaller than a matchbox, and just a bit bigger than the tip of his thumb.

“We have been working with ultracheap devices based on commercial off-the-shelf microcontrollers. This has got a tiny camera, and a visual application to take pictures and to perform object detection in a reasonable amount of time. The idea is to replace expensive purpose-built sensors with generic cameras and deep learning,” Edvardsen describes.

“Taking pictures is a sensitive topic, but one of the great things with TinyML is privacy. The pictures never leave the device but are consumed immediately. There is no need to orchestrate privacy issues”, Gard Spreemann points out.

“You can throw these out in massive numbers at a negligible cost. If these become ubiquitous, the demand of connectivity will go up. This overlaps nicely with NB-IoT,” Gard Spreemann says.

NB-IoT (Narrowband IoT) is a mobile technology made specifically to fulfill IoT requirements of low power consumption. Telenor, DNA, and other operators have extensive NB-IoT networks.

Multiple practical use cases

TinyML is optimized to minimize consumption of both energy and bandwidth. Still, the user needs to consider carefully what trade-offs can be accepted and what data should be uploaded. Data is sent only intermittently, and the device remains mostly in sleep mode.

“With TinyML you can put eyes in the real world, and these eyes can be everywhere. It is perfect when there is need to see something on the field without going out every day. You can place devices in rural areas where there is no broadband access,” Edvardsen describes.

His team has considered use cases for many different industries. Often, TinyML devices can form a part of larger systems. TinyML suits well for tracing and counting cars, buses, bikes, hikers, or even wildlife.

“Agriculture will benefit a lot from AI technology and TinyML in the coming decades, for instance, to precisely monitor crops or herds of animals, or to estimate the need of pesticides. A transportation operator can count passengers on bus stops without infrastructure,” Edvardsen says.

Many companies are actively developing TinyML software and hardware. For example, Google is developing TensorFlow Lite for Microcontrollers.

“To summarize what we have learned is that TinyML is coming, the algorithms are coming, and hardware is improving. The foundations for an IoT revolution are in place. That makes it very exciting,” Edvardsen concludes.

What is TinyML and its benefits:

  1. Very cheap and small IoT sensors
  2. Simplified machine learning for object recognition 
  3. Low power and low bandwidth consumption
  4. Flexible for multiple use cases
  5. High privacy

Telenor Group provides IoT solutions both in the Nordic countries and globally under its Telenor IoT brand. Telenor IoT combines the IoT expertise and experts of Telenor Connexion and Telenor's subsidiaries. Its offering is available through all of Telenor's Nordic country organizations, Telenor Connexion and selected partners.


IoT solutions are rapidly growing more popular. Tampere-based IGL-Technologies utilizes LTE-M technology to manage car charging posts – why did the company end up with this solution and what should anyone interested in IoT solutions know about choosing the right technology?