With these resources, your Pi can think for itself. PJ Evans gets (machine) learning
The Raspberry Pi is a powerful tool when it comes to artificial intelligence (AI) and machine learning (ML). Its processing capabilities, matched with a small form factor and low power requirements, make it a great choice for smart robotics and embedded projects. Google is a champion of the Pi’s place in the AI world, its AIY voice recognition system being given away with this very magazine (issue 57, no longer available in print).
Google’s AI Education site is an ideal place to start your machine-learning journey. If you want to really understand how AI/ML works ‘under the hood’, there are lot of principles to comprehend before you even get to coding. Google has provided a self-guided suite that starts with a ‘Crash Course’ in machine learning, then expands to cover the basics of problem framing and data gathering. The main online course comprises 25 lessons over 15 hours (approximately, you can set your own pace) and comes in the form of reading materials, interactive sections, programming exercises, and video tutorials. This is then backed by a substantial collection of follow-on courses. A superb resource.
TensorFlow is, without a doubt, the most popular software library for machine learning on the Raspberry Pi. If you’re keen to get started and write code, TensorFlow will probably be your tool of choice. You can get a great introduction to TensorFlow in The MagPi #71, but if you are after a deep dive, Udemy offers a comprehensive 14-hour video course that covers not only the theory of machine learning, but also the practicalities of setting up the software with real-world examples and programming exercises. If you’re after a hands-on learning approach, this may well suit. Don’t be put off by the steep (£195) price – this course is often promoted and was £13 at time of writing.
This book is perfectly tailored to the Raspberry Pi community. Not only does it cover the principles behind concepts such as neural networks, fuzzy logic, and shallow versus deep learning, it also provides practical, fun projects to code and build. Starting with simple examples of learning, you can play your Pi at noughts-and-crosses and Nim. Along the way, the projects are made fun through the use of the Pi’s GPIO header, using LEDs and switches to bring code to life. You then progress to robotics, covering obstacle avoidance and light seeking. A steady learning curve culminates in the building of ‘Alfie’, your very own artificially intelligent robot vehicle. If you fancy building the winner of the next Pi Wars, this could be the perfect reading material.