Intro to TinyML Part 2: Deploying a TensorFlow Lite Model to Arduino | Digi-Key Electronics

In this tutorial series, Shawn introduces the concept of Tiny Machine Learning (TinyML), which consists of running machine learning algorithms on microcontrollers.

In this episode, we create an (as simple as possible) Arduino sketch to load our TensorFlow Lite model file and run inference. The model is used to predict values of a sinewave, which we can graph using the Serial Plotter. An LED is hooked up to the Arduino to demonstrate how to connect hardware to machine learning.

TensorFlow has a pre-built library that we can install in Arduino. We use the functions from this library to load our model and run inference to make predictions. Note that at the time this video was made, TensorFlow Lite for Microcontrollers is still under heavy development, so many of the features can change on a regular basis.

Before starting, we recommend you watch the following videos:

Product Links:

Related Videos:
Intro to Edge AI

Getting Started with Machine Learning Using TensorFlow and Keras

Intro to TensorFlow Lite Part 1: Wake Word Feature Extraction

Intro to TensorFlow Lite Part 2: Speech Recognition Model Training

Low-Cost Data Acquisition (DAQ) with Arduino and Binho for Machine Learning

Intro to TinyML Part 1: Training a Neural Network for Arduino in TensorFlow

Related Articles:
What is Edge AI?

Getting Started with Machine Learning Using TensorFlow and Keras

TensorFlow Lite Tutorial Part 1: Wake Word Feature Extraction

TensorFlow Lite Tutorial Part 2: Speech Recognition Model Training

TensorFlow Lite Tutorial Part 3: Speech Recognition on Raspberry Pi

Low-Cost Data Acquisition (DAQ) with Arduino and Binho for ML

Intro to TinyML Part 1: Training a Model for Arduino in TensorFlow

[seanaf]

I programmed my computer to perform the operation of flying out my window onto the concrete sidewalk below. It performed this operation flawlessly. I'm a genius. Go me!

[billward]

Wow- so much content! I watched these vids 4x and I’m still not sure I caught everything you said. Excellent video Sean AF!

[mark]

Really cool! I paused when I saw some of that Arduino code. Delegate functions, used by reference, passed into namespaces. Developer gold for me.

[gregbushta]

That was awesomely fast. You covered a lot of ground in very little time. Now to watch it again. Thanks.

[edwardwright]

Wow. Thx man. Feels like I've devolved several hundred thousand years!

[indianasb]

Who gave up and read the comments as he was explaining this?

[donniebrookins]

I couldn't find the ENGLISH version

[basildouglas]

Sir, I found your presentation to be brilliant. By far the very best I have ever heard. You certainly have a level of understanding that you can be proud of. There is one point that I humbly request that you allow me to raise. Were you speaking in English? I never understood a word you said. But I also realize that people of your caliber are hard pressed for your time, and perhaps this may have been the reason you spoke at a speed that I have never known, it was just awesome, well done. In closing I have to admit that I never understood anything. To say that it went over my head, is an understatement.
I was so excited to have found your channel. But sadly the only info I managed to write down was “Tensor Flow Lite.” Did I miss something?

[Aswaguespack]

The surname Hymel evolves in the Settlements of the German Coast on the Right Bank of the Mississippi River above New Orleans in what is now St. Charles Parish.

The original name was “Immel”. All descendants with variations of “Hymel” come from David Immel 1712-1782.

On January 8th to Peter Immel, and Susanna, his wife was born a little boy and baptized on the 10th of the same month, named David. David was born in Stebbach, Baden Germany, baptized in the Lutheran Church.

David Hymel [Himel] was the Progeniton of all the Hymel/Himel families of Louisiana.

David arrived in Louisiana an orphan, his whole family died enroute from Germany to Louisiana. David Hymel [Himel] and Marianne Schantz were the parents of 13 Children.

The spelling of the Hymel name is indigenous to French South Louisiana. My mother was an Hymel.

You may be a relative!!

Great info!!

Anthony

[belliumm]

Shawn can you create a part 3 to this video series that shows how to run inference on the speech recognition model you created in your Intro to Tensorflow Lite series on the Arduino?

[buildingconstructioninnova]

Great video Shawn. Thank you. Please have you tried using model.h file at 490KB? Just wondering before we implement the model. All the best

[jimmy_jamesjams_a_lot]

You know, where you declared the ‘float’ing types with the ‘constexp’, that’s where I feel I need to understand just what’s being done at hardware level. How much longer would it take to use doubles? How can the compiler help me in that, besides running code and then reviewing counted clock cycles? How can I use that pwm output to make an LFO audio effect, like tremolo? Please keep this up as I’m certain you can throw us some even more down to real world applications?! THANKS A BUNCH!! THIS IS GREAT!!

[eb-l]

I have no idea why this was recommended to me. I just watch memes :/

[alkmst-xyz]

Any hope to train on training a convolutional network?

[TheRainHarvester]

Interesting, but what other things could be done? Camera data would be too large - right?

[syaduinotech]

Very advance. Tq sir. Do you mind to share the example of input and output. Based on model as your example.

[theunderjack]

It’s videos like this that make me understand I know nothing

[SimoneIovane]

What if my input is a time series? How should I pass it to the model input?

[blobanater]

Hi Shawn, thanks for the video! I'm wondering how you would set the 'model_input' to something with greater than 1 dimension. The input dimensions for my model follows the shape (1, 45, 45, 1) for an image input. Thanks.

[delawarepilot]

Very cool. I need to get into tensor flow. So many possibilities.