Chapter 21. Learning More
We hope that this book helps you to solve problems that matter, using inexpensive, low-power devices. This is a new and rapidly growing field, so what we’ve included here is just a snapshot. If you want to stay up to date, here are some resources we recommend.
The TinyML Foundation
The TinyML Summit is an annual conference that brings together embedded hardware, software, and machine learning practitioners to discuss collaborations across these disciplines. There are also monthly meetups in the Bay Area and Austin, TX, with more locations expected in the future. You can check the TinyML Foundation website for videos, slides, and other materials from the events, even if you can’t make it in person.
SIG Micro
This book focuses on TensorFlow Lite for Microcontrollers, and if you’re interested in contributing to the framework there’s a Special Interest Group (SIG) that enables external developers to collaborate on improvements. SIG Micro has public monthly video meetings, a mailing list, and a Gitter chat room. If you have an idea or a request for a new feature in the library, this is a good place to discuss it. You’ll see all the developers working on the project, both inside and outside Google, sharing roadmaps and plans for upcoming work. The usual process for any changes is to start by sharing a design document, which can be just a single page for simple changes, covering why the change is needed and what it will do. We usually publish this as an RFC (“request for comment”) to allow stakeholders to contribute their feedback, and then follow it up with a pull request containing the actual code changes once the approach is agreed.
The TensorFlow Website
The main TensorFlow website has a home page for our work on microcontrollers, and you can check there for the latest examples and documentation. In particular, we’ll be continuing our migration to TensorFlow 2.0 in our training sample code, so it’s worth taking a look if you’re having compatibility problems.
Other Frameworks
We’ve focused on the TensorFlow ecosystem given that this is the library we know best, but there’s a lot of interesting work happening on other frameworks, too. We’re big fans of Neil Tan’s pioneering work on uTensor, which has a lot of interesting experiments with code generation from TensorFlow models. Microsoft’s Embedded Learning Library supports a large variety of different machine learning algorithms beyond deep neural networks, and is aimed at Arduino and micro:bit platforms.
Have you built an embedded machine learning project that you’d like to tell the world about? We’d love to see what problems you’re solving, and one great way of reaching us is by sharing a link on Twitter using the #tinyml hashtag. We’re both on Twitter ourselves as @petewarden and @dansitu, and we’ll be posting updates on this book at @tinymlbook.
Friends of TinyML
There are a lot of interesting companies working in this space, from early-stage startups to large corporations. If you’re building a product, you’ll want to explore what they have to offer, so here’s an alphabetical list of some of the organizations we’ve worked with:
Wrapping Up
Thanks for joining us on this exploration of machine learning on embedded devices. We hope that we’ve inspired you to work on your own projects, and we can’t wait to see what you build, and how you can drive this exciting new field forward!