TinyML Cookbook: Combine artificial intelligence and ultra-low-power embedded devices to make the world smarter

The book TinyML has a wealth of great learning projects for the hands on engineer and data scientist. It's not your typical run of the mill data science book, it's an indept journey through projects into TinyML and deployment of example projects. The book has an intimidating 665 pages, however you can just start with one of the many projects. I would recommend to read chapters one and two as they will teach you the basic and skills you need for each project.Every project start with a list of prerequisites and optional tools, e.g. book snippet ### To complete all the practical recipes of this chapter, we will need the following:• An Arduino Nano 33 BLE Sense• A Raspberry Pi Pico• A SparkFun RedBoard Artemis Nano (optional)• A micro-USB data cable• A USB-C data cable (optional)• 1 x half-size solderless breadboard (Raspberry Pi Pico only)• 1 x AM2302 module with the DHT22 sensor (Raspberry Pi Pico only)• 3 x jumper wires (Raspberry Pi Pico only)• A laptop/PC with either Linux, macOS, or Windows. ###Then an overview of the different steps, so you have a good idea of your 'mini' milestones. Then each chapter explains how to reach these milestones, it includes code and images, links to webpages and data downloads, etc. The very first project is already good fun, it's how to bring your own weather station alive!If you are interested in practical engineering using ML, this is the book...it's a must have!

Background: I've been involved with TinyML for over 6 years through working at Qeexo, and was at the inaugural TinyML Meetup at Qualcomm in the Silicon Valley. I also read Pete and Daniel's "TinyML: Machine Learning with TensorFlow Lite".That said, I found the TinyML Cookbook... well, very "cookbook"-like. Besides a pretty general (but good) intro of TinyML not unlike my own TinyML pitch, it comes with very detailed, step-by-step instructions and pre-prepared code on GitHub, so it's super easy to get started, even if you're not a coder or embedded engineer. (I am neither, but I do have a CS degree from a long time ago.)This Cookbook is very well-structured, and broken down into individual projects and categories. One can select the tools and hardware one is interested in, and start there, without having to go through the entire book, similar to how you can pick out one dish to cook from a cookbook. Full disclosure: I didn't do every single project, just selected a couple that I already have the hardware for (Arduino Nano 33 BLE Sense, Raspberry Pi Pico).Chapter 8 on the Arm microNPU is actually the most interesting to me as the latest and greatest tech in TinyML, however:"Arm Ethos-U55 is the first microNPU designed by Arm to extend the ML capabilitiesof Cortex-M-based microcontrollers. Unfortunately, there is no hardware availabilitywith this new processor at the time of writing. However, Arm offers a free Fixed VirtualPlatform (FVP) based on the Arm Corstone-300 system to quickly experiment with MLmodels on this processor without the need for physical devices."So, I was not able to test with actual hardware.Overall, it was quite fun to try some new things out. One thing I wish it had is more room for experimenting or expanding upon the current project - pointers on how to change the current code to achieve different results, or other applications and real-life use cases of these "demo" projects. (However, I understand that's not really in the scope of a "cookbook" - one usually doesn't just experiment with random ingredients to see how a dish turns out differently.)Fun read!Anyway,

It's been a while from Pete Warden's 'TinyML', and we finally have another reference book. Overall, it is a great introductory book for people who want to do some practical exercises with machine learning on microcontrollers. Although the author says it's for ML engineers interested in MCU in "Who this book is for" part, I would say it's also for embedded engineers who want to learn some ML stuff.The recipes cover most of the common and cutting-edge use cases, e.g. keyword spotting and image classification. All the important concepts and tools are discussed, such as what is quantization, what are TensorFlow Lite and TensorFlow Lite for Microcontroller, and how they work, how to use EdgeImpulse etc. The last chapter even shows how to use ARM's virtual hardware to develop with Ethos-U55 NPU.Some very detailed information is also given in between the lines. For example, I really like the explanation of the RGB565 to RGB88 conversion is chapter 5, which is clear and explicit. Besides, as long as some external sensor is needed, all of specs are provided. If you need to build a circuit on breadboard, the step-by-step illustration is very easy to follow.Unfortunately, I don't have the time to do some hands-on practice. But if you want, all the links on GitHub are given in the book. For most of the recipes, you don't need to download the dependencies, as the author uses Colab notebook.What I really wish the author could have added is how the readers can expand on the recipe, maybe just list some ideas to let the readers explore by themselves. Another thing is, in some recipes you need to build a neural network, and the author gives all the parameters. It would be better to explain a little bit why such architecture is selected, also maybe provide a range (like number of neurons) for readers to tweak the network.In conclusion, I will definitely have a hard copy of this book on my shelf. No matter you are ML engineer want to develop something on MCU, or embedded engineer want to develop something about ML, this book is a good starting point.

When I got this book, I thought it would be like a tiny reference guide for ML applications, lo and behold, I got happily enticed into the world of IoT devices and how to get around their processing limitations to implement Machine Learning in them. A happy surprise and now one of my favorite books.