uTensor – Embedded Devices And Machine Learning Models

TensorFlow has already made it possible to use trained machine learning models on smartphones. TensorFlow Lite goes one step further and run TensorFlow models on a Raspberry Pi.

uTensor even puts AI on a microcontroller (MCU). They are small and cheap, but they are also energy efficient, slow and have little RAM, which doesn’t make it any easier.

In my presentation I will take a simple machine learning model on TensorFlow and show why it will NOT work with uTensor. But don’t worry, I will also show a working example. I will also go a little bit deeper into TensorFlow operators and you will see also some C/C++ code.

Objective of the talk

The participant will learn some TensorFlow operators, how to save a model and convert it into C/C++ code, something about uTensor, and why a MNIST example is easier to run on uTensor than a XOR model.

Required audience experience

Programming skills and TensorFlow knowledge are an advantage, but not mandatory

Track 3
Location: Burton and Redgrave Date: September 30, 2019 Time: 2:00 pm – 2:45 pm Lars Gregori, SAP CX