Why TinyML use cases are taking off
TinyML technology can successfully collect and analyze data in real scenarios, as demonstrated in various use cases.
Machine learning is commonly associated with big data, but that's changing quickly. While IoT, edge computing and intelligent edge devices arguably make big data even bigger, not all data at the edge is useful. Therefore, the data needs to be analyzed at the edge to separate the signal from the noise.
Other types of pattern recognition are also occurring at the edge, such as the discovery of patterns in data depicting the impact of weather events on crops or people at risk of heart attacks. Determining the status of something in the field requires machine learning. Yet, IoT devices tend to be low-power devices. Enter TinyML and endless use cases for edge analytics become possible.
"TinyML is already being used for speech and image recognition, neural networks and other applications," said Muhammad Taimoor, data science trainee at training platform provider Data Science Dojo. "It's a simple, intuitive machine learning API and also a great starting point for anyone interested in learning more about machine learning."
Google's TensorFlow Lite is the most popular TinyML open source deep learning framework. It enables developers and data scientists to deploy machine learning models on mobile and IoT devices.
There are several examples of current and potential TinyML use cases.
Loftera, a price comparison and home improvement advice website for U.K. households, plans to use TinyML for predictive maintenance on construction machines.
"It is possible to save significant amounts of money by using predictive maintenance because machines are prone to failing," said Zac Houghton, CEO of Loftera.
Swedish edge AI product company Imagimob is working with an EU project involving 55 organizations throughout Europe to understand how TinyML and cloud services can support companies in crop management and livestock use cases.
For a crop management demonstration, Imagimob outfitted two tractors with Dialog IoT Kit (Bosch sensor) devices and mounted Android phones on their dashboard holders. Then, the data would either be labeled in real time by the operator or in post-production by viewing a smartphone video stream. Using Imagimob's AI software, a number of neural networks were trained and deployed together with sensors, batteries and a long range (LoRa) radio. This allowed the farmer to monitor mobile assets using data from the accelerometer and gyroscope sensors. The end result was sent over a LoRa network periodically in near real time.
Imagimob has also used the same tools to monitor livestock health and wellbeing as a proof of concept.
The developers who built the FindPeopleFast.net site are going into the grocery store monitoring business, according to CEO Daniela Sawyer.
"Supermarkets often struggle to keep track of goods on shelves. This is currently done by humans manually, and glitches commonly occur," said Sawyer. "Our system will consist of a camera, predefined information about goods being monitored and the amounts that should be present. As soon as the goods need replenishing, the system will send alerts."
Data privacy and security
It might be better to analyze data locally for data privacy and security. Examples include clinical trial or healthcare data gathered from wearables. Specifically, TinyML can be used for:
- monitoring the vital signs of high-risk individuals;
- predictive screening;
- patient monitoring following a hospital visit; and
- tracking the trajectory of a degenerative health condition.
Deep learning models tend to require a lot of computing power, which increases a company's carbon footprint. While TinyML won't replace those models, it does provide a more sustainable way to analyze the glut of IoT data at the edge.
"Making TinyML more accessible to developers will be critical to transforming waste data into actionable insights and creating new applications across a wide range of industries," said Hassan Usmani, tech expert at smart lighting company Yeelight. "With new types of human-machine interfaces emerging and the number of intelligent devices growing, TinyML has the potential to make AI and computing at the edge ubiquitous, cheaper, scalable and more predictable."
Hassan UsmaniTech expert, Yeelight
Hyper-personalization is the latest customer experience trend. To enable it, companies must take what they know about customers from their core and mission-critical systems and marry that with users' context, as appropriate, at any given time. Accomplishing that requires TinyML at the edge to understand customers' context, including their behavior.
A similar use case exists for personal safety. For example, if a pedestrian walking down the street suddenly has an adrenaline rush, screams and starts running, the person's GPS coordinates and behavioral data could be sent to an emergency response team. This team will then verify the data by cross-referencing it with other available data, such as video footage or a call made by a witness.
More organizations across many different industries and verticals will adopt TinyML to help analyze data on low-power devices, in real time, at the edge.