TinyML Cookbook: Combine machine learning with microcontrollers to solve real-world problems , Second Edition

Having read the first edition of this book that I own, I received a pre-release copy of the second edition from <PACKT> to review for this book. I was a co-editor on another <PACKT> book related to RTOS (Real-time Operating Systems) so I get pre-release copies from time to time to review.This book is a great expansion of the first edition and includes more visual diagrams and expanded detail to explain hardware connectivity, MEMS sensors and how they operate, different types of machine learning inference with sensor devices, the Edge Impulse cloud-based no-code machine learning toolkit, and Tensorflow programming using the Arduino IDE.I wouldn't consider this book for absolute beginners but a beginner would need to read it a couple of times first to understand core concepts before trying to do the "How To Do It" sections at the end of each example project. This book is more suited with someone who has some exposure to embedded microcontroller programming with Arduino IDE, Arduino dev boards like the Nano 33 BLE Sense, the Raspberry Pi Pico dev board, and perhaps the ESP32 dev board variants from Espressif Systems.The new Arduino Nano 33 BLE Sense 2 has recently come out and should apply to this book as well for the Edge Impulse and Tensorflow chapters for deploying TinyML machine learning models. If you buy this book now and buy an Arduino Nano 33 BLE Sense dev board and peripherals for Christmas, you can have enough time to read the book and deploy TinyML models over the Christmas holidays after your dev board arrives!I work with embedded machine learning on intelligent wireless IoT devices for my business and can deploy TinyML models to almost any ARM Cortex-M embedded microcontroller out there. I use other machine learning tools to deploy TinyML models directly onto MEMS sensors as well.Gian Marco Iodice is an expert in the field of embedded machine learning due to his work at ARM in the UK and his education experience in researching the field of TinyML on embedded systems or resource-constrained embedded devices for computer vision. The principles of this book cover a wide range of TinyML possibilities with great examples from deploying machine learning models from scratch using the Arduino IDE with C and C++ code and ARM MBED OS to no-code tools like Edge Impulse.For anyone wanting to learn how to deploy machine learning models to an embedded microcontroller development kit like the Arduino Nano 33 Ble Sense or the Raspberry Pi Pico dev kit, you must get this book to learn how to do it easily while learning important concepts at the same time. You can also join the "Embedded Systems Professionals" Discord channel to ask the author of the book, Gian Mardo Iodice, questions about the contents of the book and to get some help on how to deploy TinyML models to your dev board.To conclude, I know you will enjoy the book as much as I did. The second edition is an improvement to the first edition with updated code fixes, more diagrams, expanded explanations of topics, and updated information. Buy an Arduino Nano 33 BLE Sense dev board, buy some peripheral sensors to connect to your Arduino dev board, and start deploying TinyML models with the "TinyML Cookbook: Combine machine learning with microcontrollers to solve real-world problems" today. I highly recommend this book if you want to learn about the future of machine learning on embedded devices and how to actually deploy TinyML models onto embedded systems to make those systems really smart.

If you're into microcontrollers and machine learning like I am, this book is an absolute gem. It's the perfect fusion of both worlds, providing valuable insights on how to leverage machine learning to tackle real-world challenges on power and compute-constrained devices.

I’ve spent much of the past 7+ years working with various groups of software developers focused on the problem of 'cramming' AI/ML capabilities into various microcontrollers and microprocessors found in all sorts of resource-constrained devices. These ranged from tail tags on cows, gait monitors for Parkinson patients, engines on large scale mining equipment, and video cameras on mini-drones and low earth orbit satellites, to name just a few. During each of these projects a sizeable percentage of our time was spent gathering and assessing various books, papers, research papers and the like in order to address the problem. This is how I came across Gian Marco Iodice’s ‘TinyML Cookbook – both the the 1st edition (April 2022), and now the 2nd edition (November 2023) . If these editions had been published earlier, then they could have served as a valuable resource, saving us time, reducing our false starts, and introducing much more rigor to our development processes.Like its predecessor, this 2nd edition of the Cookbook has a decidely practical hands-on approach (much like the ‘Make’ books for microcontrollers). Starting with an intro to TinyML, ML, and deep learning (DL), it quickly guides readers through the steps required to setup the microcontroller hardware and software environments used to ‘cook’ the recipes in the remainder of the book. Most recipes in this Cookbook employ the Arduino Nano 33 BLE Sense, the Raspberry Pi Pico, and occasionally the SparkFun Artemis Nano. Likewise, the same dev environment (Arduino IDE or Arduino Web, TensorFlow, Edge Impulse, and TensorFlow Lite) and development procedures are in most of the chapters, taking the reader from initial project statement to prototype completion. Crudely put, for each project the reader starts on the cloud with the goal of building, training and testing the appropriate TensorFlow ML model. Once completed, TensorFlow Lite and quantization are used to create an ML model that can then be deployed on the various microprocessors of interest.What makes this Cookbook enjoyable is the non-trivial nature of its projects. While some may seem basic at first glance, such as predicting snowfall from sensor data or controlling LEDs with voice commands, diving into the details of these, as well as the other projects, reveals their complexity. These projects can serve as building blocks for more intricate applications.In the world of Amazon reviews, we often scrutinize 1s and 2s in deciding whether to purchase the product. For the earlier edition, there were no such ratings, just one 4 and the rest 5s (28 of them), with reviewers praising its practicality, readability, and comprehensive coverage. From my own experience, these accolades extend to the 2nd edition.Who should consider this book? It's not for non-programmers or beginners. Instead, it's ideal for those with ML and programming experience in Python and C/C++ who want to explore tinyML and Edge AI. Conversely, if you're comfortable with microcontrollers and programming but new to ML, this book is an excellent starting point. If you're well-versed in both ML and microcontrollers and seek a deeper understanding of tinyML, this book is a valuable addition to your library. Finally, if you already have the 1st edition, then the upgrade to the new edition is well worth it.

This book is an invaluable resource for both enthusiasts and professionals interested in integrating machine learning with microcontrollers. This practical guide, focusing on tinyML, is perfect for those with a basic understanding of machine learning and an interest in applying it on microcontrollers like Arduino Nano 33 BLE Sense, Raspberry Pi Pico, and SparkFun RedBoard Artemis Nano.This book is a comprehensive journey into tinyML, starting with fundamental concepts and advancing to real-world sensor applications. It covers a wide range of topics including weather prediction, voice-controlled LEDs, music genre recognition, object detection, and gesture-based interfaces for YouTube playback. Practical projects using Edge Impulse, TensorFlow Lite for Microcontrollers, and Apache TVM are included, making it an essential read for those looking to explore tinyML applications.This edition of the book stands out with its hands-on approach and detailed recipes that guide the reader through various tinyML implementations. It's particularly useful for readers looking to apply machine learning in memory-constrained environments, offering insights into model optimization and deployment strategies.Whether you're a beginner or looking to expand your ML knowledge, this book's real-world examples and end-to-end project guidance make it a must-read for anyone venturing into the field of tinyML.

TinyML is the set of technologies to enable using Machine Learning (ML) on microcontrollers (MCU’s) for embedded systems. My background is in embedded systems, but I’m a novice at ML.The book is excellent. It’s not a book on theory. Instead, it’s a thorough, practical hands-on guide to applying ML on embedded systems. It assumes no background knowledge in ML or working with MCU’s, providing enough detail to familiarize the reader with the concepts and terminology needed to execute the recipes. It provides a number of links and references for more information.ML is a complex topic, and applying it on MCU’s adds further complexity due to their constraints of limited CPU performance, limited electrical power, limited memory, and lack of floating point hardware. Making it work requires a combination of cloud, local host, and embedded system MCU on-device elements.The book starts with an overview of ML technologies, followed by an excellent brief introduction to working with MCU’s. Readers interested in learning more about MCU’s and the various devices that form embedded systems can use one of the many educational electronics sets for Arduino and Raspberry Pi.The recipes cover using ML on three different MCU development platforms: Arduino Nano 33 BLE Sense, Raspberry Pi Pico, and SparkFun RedBoard Artemis Nano. These provide a variety of on-board sensors and connections for external sensors.The book uses the Arduino and RPi as the primary platforms for detailed recipes, and the Artemis for additional exploration. In addition, it shows how to run ML on the QEMU emulator on your local host, which allows you to experiment with more platforms. This all shows you how to work with a variety of systems.One of the main challenges for the beginner is the bewildering array of software tools that need to be used for tinyML. The real strength of the book is showing how to do that.The recipes provide detailed instructions for setting up and using MCU development environments (Integrated Development Environment (IDE) and Command Line Interface (CLI)), CMSIS-DSP, MbedOS and Zephyr OS, QEMU, and the ML tools TensorFlow Lite, Edge Impulse, Apache TVM, and scikit-learn. They also include custom code snippets in C++ and Python for integrating and adapting items. They impart a wealth of practical knowledge that can be applied beyond the book.The projects covered by the recipes: a weather station that interprets temperature and humidity sensor data to predict snow; controlling LED’s with voice commands received over a microphone; recognizing music genres on the mic; detecting objects in camera images; and interpreting gestures with an Inertial Measurement Unit (IMU, i.e. accelerometer and gyroscope) to control YouTube playback.