TinyML Frequently Asked Questions
TinyML Frequently Asked Questions
Q: What is TinyML?
A: TinyML is a branch of machine learning that uses small form factor devices attached with sensors including bare metal or RTOS based microcontrollers to execute ML models and algorithms as software.
Good things, like TinyML, come in small packages.
Another way to look at TinyML technology is as a smaller subset of Machine Learning & Deep Learning technology including software, models, algorithms, applications and devices.
Q: Is TinyML same as Embedded ML?
A: TinyML is implemented of a subset of embedded devices with power lower than 1 Watt. TinyML devices mostly run on AA or coin-cell battery, whereas most embedded devices may use Li-Ion battery or plugged in an electric outlet. Because of power restriction, TinyML devices tends to be bare metal devices. TinyML devices do not have enough resources like hard disk (flash) and RAM to load an operating system. Some TinyML devices may have a variation of RTOS (Real Time OS), whereas embedded Linux is a popular choice for OS among plugged embedded boards.
Not all embedded devices are TinyML devices, but all TinyML devices are embedded devices.
TinyML vs Embedded ML
Q: What is a TinyML device?
A: TinyML devices include Microcontrollers and in some cases Programmable logic controllers (PLCs). These controllers are ubiquitous with low-power and compact form factor that allows them to be used as edge devices for low-latency, low-power, and low-bandwidth model inference. Current count of these devices exceed over 300 billion globally.
Future of ML is like a Blackhole - Tiny & Mighty
In contrast to a typical CPU's power consumption of 65 to 85 watts and a typical GPU's power consumption of 200 to 500 watts, a typical microcontroller uses power on the order of milliwatts or microwatts. This is over a thousand time power savings and a thousand times less energy. Due to their low power consumption, tinyML devices can run ML applications on the edge while remaining unplugged from internet or electricity for weeks, months, and in some cases, even years.
Q: What is TinyML used for?
A: TinyML is popular among applications that run in remote areas with connectivity contested or denied environments like deep space, defense, farming and others. Here are the 4 popular applications on TinyML mentioned in the book - Introduction to TinyML.
Audio Wake Word Detection
Predictive Maintenance
Visual Wake Word Detection
American Sign Language
There is a list of over 130 TinyML projects as part of verticals in AITS Cainvas.
Q: What are the popular applications of TinyML?
A: There are over a thousand applications that have gained popularity in over 20 verticals including.
Smart Industrial IoT
Smart Aviation
Smart Farming
Smart Transportation and Logistics
Smart Security
Smart Oil and Gas
Smart Environment
Smart Space
Smart Home
Smart City
Smart Retail
Smart Energy
Smart Auto
Smart Society
Smart Finance
Smart Health
A significant number of of verticals are listed in AITS Cainvas.
Q: What are the best TinyML books?
A: TinyML is a relatively new field, so amount of literature is still evolving. Here is what is available so far for TinyML projects:
Non-Tech and Beginner’s No Code TinyML Book - Introduction to TinyML (Price: $0 to $2.99)
Data Scientist TinyML Book - Machine Learning with TensorFlow Lite on Arduino (Price $30.99)
Embedded Developer TinyML Book - TinyML Cookbook for embedded devices (Price $36.99)
Q: Is TinyML a open source framework?
A: TinyML is a foundational technology term (similar to embedded) used as an extension to describe other technologies like tinyML devices, tinyML boards, timyML software, tinyML compiler etc. TinyML compilers and interpreters like AITS deepC compiler and Google's TensorFlow Lite Micro are open source.
Q: How can I learn TinyML?
A: Most of the courses and information on TinyML free. Here is a free TinyML course for beginners and intermediate learners.
Q: What is a TinyML device?
A: TinyML devices include Microcontrollers and in some cases Programmable logic controllers (PLCs). These controllers are ubiquitous with low-power and compact form factor that allows them to be used as edge devices for low-latency, low-power, and low-bandwidth model inference. Current count of these devices exceed over 300 billion globally.
Future of ML is like a Blackhole - Tiny & Mighty
In contrast to a typical CPU's power consumption of 65 to 85 watts and a typical GPU's power consumption of 200 to 500 watts, a typical microcontroller uses power on the order of milliwatts or microwatts. This is over a thousand time power savings and a thousand times less energy. Due to their low power consumption, tinyML devices can run ML applications on the edge while remaining unplugged from internet or electricity for weeks, months, and in some cases, even years.
If you're a human, send your questions to tinymlbook "at-the-rate-of gma!l bot calm"