Fraunhofer IMS and ARDUINO® announce the release of AIFES, a standalone, high-efficiency, AI framework for microcontrollers.
Fraunhofer IMS and Arduino announce the release of AIfES, a standalone, high-efficiency, AI framework completely programmed in C, which allows innovators to train and run machine learning algorithms even on the smallest microcontrollers.
The high optimization of the framework allows even the 8-bit controller of the Arduino Uno to implement an ANN that can be trained in moderate time. This enables the creation of customizable devices that can adapt to the task at hand by means of training without using an external computer.
Small self-learning battery-powered devices can be developed that are independent of a cloud or other devices. Sensor data can be processed where they are generated, directly in the device. Training data can be captured directly in the device and used for training.
The Fraunhofer IMS with AIfES and Arduino prepare to enter a partnership that makes it possible to integrate AIfES directly into the Arduino IDEs through the Arduino library manager.
AIfES is comparable to and compatible with the well-known Python ML frameworks such as TensorFlow, Keras or PyTorch, (currently with less functionality although new features are constantly added). The current version supports feedforward neural networks (FNN), which can be completely configured freely. It also already integrates common activation functions like ReLU, sigmoid or softmax. A full implementation of Convolutional Neural Networks (ConvNet) will follow in the near future.
It is possible to import a trained ANN from another ML framework. Nothing but the network structure and weights are required to map an ANN and even train it further. A Keras example is included in the library and it is also not necessary to port the model to TensorFlow Lite.
All algorithms are optimized for use on resource-limited embedded systems and even the required memory area for the ANN can be specified by the developer. AIfES has been developed in a modular way to be able to exchange different components of the algorithms, e.g. a matrix multiplication.
The Fraunhofer IMS has been using AIfES internally in AI research and development for years. AIfES has proved effective as a development tool for customer-specific AI solutions, and has been used in several interesting demonstrators, included in many public projects, and integrated into a number of future commercial products. Demonstrators include an extremely compact handwriting recognition system realized on an Arduino Uno, a wireless current sensor for condition monitoring and a gesture recognition system.
AIfES is offered within a dual license model. It can be used free of charge for open-source software under the GNU General Public License (GPL) version 3. For commercial products, a license agreement must be signed with the Fraunhofer IMS.
(please contact the AIfES team for details).
Dr.-Ing. Pierre Gembaczka (Inventor / Product Manager of AIfES & “Industrial AI” Program Manager): “AIfES started in 2016 as a small project of mine to explore self-learning sensors. I never thought that I would be able to introduce a whole framework together with Arduino. The Arduino Uno was the first board I tested AIfES on, so it’s just a perfect fit. The computing power of the new Arduino Pro series will bring the possibilities to a whole new level. Being a maker myself, I’m incredibly excited to be able to share AIfES with the community to create exciting DIY projects”.
Prof. Dr. rer. nat. Anton Grabmaier (Director of the Fraunhofer Institute for Microelectronic Circuits and Systems): “We are looking forward to working together with Arduino. With the integration of AI capability into the Arduino processor family there are completely new possibilities for many products. It is a significant step for AI in the product, also called embedded AI or AI of things. This step will make many everyday products into intelligent and adaptive products”.
Dipl.-Ing. Burkhard Heidemann (Head of Core Competence “Embedded Software and Artificial Intelligence”): “I am very pleased that we have succeeded in turning a small project into something really big. The hurdle of bringing AI to embedded systems has now fallen, opening up completely new scenarios for new innovative products. Through the collaboration with Arduino, everyone will now have the opportunity to enter this interesting field to implement their creative ideas”.
Justus Viga (AIfES Developer): “Quickly building a remote control with individual gesture recognition or a toaster with voice input – all of this is now possible with AIfES, and made easier than ever before in combination with Arduino. When I first got involved in artificial intelligence, I would never have believed that this could all be done even on the smallest microprocessors. The idea that with AIfES the simplest devices can learn and adapt to our needs is just incredible and I am happy to be able to work on it”.
Massimo Banzi (Arduino co-founder): “I believe this is a significant step for Machine Learning on MCUs: finally a framework designed from the ground up for constrained devices that delivers maximum performance while optimising power consumption. I can’t wait to see what people will develop with this”.
For more information, please visit: https://aifes.ai
Arduino is an open-source hardware, software, and content platform with a worldwide community of around 30 million active users. It has powered thousands of projects, from everyday objects to satellites and complex scientific instruments, providing the building blocks for innovators to power connected experiences for the future. This success has been made possible by combining a wide variety of electronic boards, easy-to-use tools, a collaborative community, and practical project examples to suit all levels.
ABOUT FRAUNHOFER IMS
The Fraunhofer Institute for Microelectronic Circuits and Systems (IMS) is one of the 75 independent institutes of the Fraunhofer-Gesellschaft for the promotion of applied research, Europe’s leading organization for industry-related research.