What is TinyML?

Whether you think it’s slightly embarrassing or downright convenient, it’s pretty neat when Netflix serves up recommendations that you’re ready to click on and binge. This and so many functions in our day-to-day lives are made possible through machine learning, or the process by which computers employ algorithms and statistics to identify patterns in large amounts of data. Maybe you’re venturing out into the world of online dating during the pandemic or staying in touch with friends and family through social media. When you’re matched in a dating platform or see updates from people you care about or are interested in the most, machine learning is at play.

Of course, as with any technology, there are upsides and downsides to machine learning, such as the need for large amounts of energy and high bandwidth, as well as lag times in data processing. The good news is that Tiny Machine Learning (TinyML) sidesteps all of those downsides while tacking on data privacy as the cherry on top.

TinyML and its upsides

Let’s take a closer look at what exactly makes TinyML tick, as well as the major upsides.

TinyML is a technique or field of study in machine learning and embedded systems[1] that explores which machine-learning applications (once reduced, optimized and integrated)[2] can be run on devices as small as microcontrollers.

The upsides to TinyML are pretty clear and compelling:

  • Quick Data Processing: Because the reduced, optimized machine-learning applications in TinyML run on edge computing, data processing is efficient. Only necessary or desired data, if any at all, is sent to long-distance cloud servers for storage.
  • Low Internet Bandwidth: TinyML requires low bandwidth since data is less frequently sent to long-distance cloud servers.
  • Low Power Requirements: Microcontrollers are the vehicles for TinyML models and have low energy requirements, which reduce charging length and frequency.
  • Data Privacy: As the so-called “cherry on top,” TinyML applications run on the edge and either do not store data on long-distance cloud servers or only send select data back to the servers

 

Getting started with TinyML

 

Early adoption & uses

Pete Warden (TensorFlow Lite Micro), Kwabena Agyeman (Arm Innovator), and Daniel Situnayake (Edge Impulse) are recognized as early influencers and “founding fathers” of TinyML.[3] Because machine learning is so widely used in the Internet-of-things (IoT) and in small, portable devices, it’s no wonder the development of TinyML happened as quickly as it did and garnered so much attention and early adoption. In 2030, ABI Research predicts the shipment of approximately 2.5 billion devices that feature TinyML.[4] And, in just the next five years, Silent Intelligence forecasts that TinyML could “reach more than $70 billion in economic value.”[5]

Uses for TinyML are far-ranging and address convenience, communication, knowledge-sharing, matchmaking, entertainment, agriculture, predictive maintenance, and healthcare, to name a few. Today, the most common fields for TinyML application include audio analytics, pattern recognition, and voice human-machine interfaces.”[6] Here are just a few of the many compelling ways TinyML can improve processes, reduce costs, and increase the quality of life:[7]

  • Predictive maintenance in manufacturing and other industries through the use of low-power sensors on equipment and machinery
  • Building automation in lighting, HVAC, and other applications
  • Vision, motion and gesture recognition in toys and entertainment
  • Pharmaceutical development and testing
  • Audio analytics in child and elderly care
  • Identifying and preventing the spread of illness in healthcare
  • Assessing the status of agricultural crops in farming

[1] Arun, “An Introduction to TinyML,” “Machine Learning Meets Embedded Systems,” Towards Data Science, November 10, 2020, https://towardsdatascience.com/an-introduction-to-tinyml-4617f314aa79.

[2] Ribeiro, Jair, “What is TinyML, and why does it matter?”, Towards Data Science, December 22, 2020, https://towardsdatascience.com/what-is-tinyml-and-why-does-it-matter-f5b164766876.

[3] Grande, Alessandro, “TinyML Enables Smallest Endpoint AI Devices,” arm Blueprint, September 24, 2020, https://www.arm.com/blogs/blueprint/tinyml.

[4] “Global Shipments of TinyML Devices to Reach 2.5 Billion by 2030,” ABI Research, September 3, 2020, https://www.prnewswire.com/news-releases/global-shipments-of-tinyml-devices-to-reach-2-5-billion-by-2030--301123076.html.

[5] Ribeiro, Jair, “What is TinyML, and why does it matter?”, Towards Data Science, December 22, 2020, https://towardsdatascience.com/what-is-tinyml-and-why-does-it-matter-f5b164766876.  

[6] Bagnoli, Margot, “TinyML: Making Smart Devices Tinier than Ever,” Plug and Play, November 4, 2020, https://www.plugandplaytechcenter.com/resources/tinyml-making-smart-devices-tinier-ever/.

[7] Bagnoli, Margot, “TinyML: Making Smart Devices Tinier than Ever,” Plug and Play, November 4, 2020, https://www.plugandplaytechcenter.com/resources/tinyml-making-smart-devices-tinier-ever/.

RELATED: 

What is machine learning?

What is artificial intelligence (AI)?

What is deep learning?