Artificial Intelligence for Embedded Systems

Machine Learning for Embedded Systems with AIfES^®

With the open source AI software framework AIfES (Artificial Intelligence for Embedded Systems) you can run and even train artificial neural networks (ANN) on virtually any hardware. Tiny embedded systems plus artificial intelligence (AI) - the topic of our time.

Open Source AI Framework

The first open source AI framework "Made in Germany", developed as a Maker project at the Fraunhofer Institute for Microelectronic Circuits and Systems IMS. AIfES^® is comparable and compatible with well-known Python ML frameworks like TensorFlow, Keras or PyTorch. In the current version, Feedforward Neural Networks (FNN) are supported, which can be configured completely freely. Common activation functions such as ReLU, Sigmoid or Softmax are also already integrated. A full implementation of Convolutional Neural Networks (ConvNet) will follow in the near future.

Tiny ML

AIfES^® can be used on almost any system, be it a microcontroller, IoT device, Raspberry PI, PC or smartphone, making the purchase of new hardware redundant. However, the focus is particularly on running AI on simple microcontrollers and small IoT devices, so-called »tinyML«. Tiny, self-learning, battery-powered devices can process sensor data where it occurs, independent of a cloud or other devices. The data is stored on the device and processing takes place without transmission delay, with significantly lower energy consumption compared to a PC.

Logo AIfES

Arduino Uno — © Wokwi
Arduino Mega Demonstrator über Wokwi

AIfES^®for Arduino^®

A version compatible with the Arduino^® IDE of AIfES^® has been realized, which can be run on almost any Arduino board.

The Repository where you can download AIfES
The Converter to create AIfES models for direct use in the Arduino IDE or other IDEs
A playful tic-tac-toe-demonstrator

Climate protection

The project supports the UN sustainability goal of climate action in industry. Through efficient algorithms and energy saving hardware, AIfES enables to drastically reduce CO₂ emissions compared to Deep Learning on high-performance computers.

Decentralized AI

AIfES enables decentralization of computing power, for example, by allowing small intelligent embedded systems to take over the data before processing it without transmission delay and provide the results to a higher-level system. This significantly reduces the amount of data to be transmitted. In addition, a network of small adaptive systems that divide tasks among themselves is also possible. A decentralized architecture allows for increased reliability and personalization.

Privacy

Since processing can take place offline on the device, no sensitive data needs to be transferred.

News are regularly communicated on the AIfES LinkedIn channel. New demonstrators and AIfES at trade fairs and events are also presented on Youtube.

Embedded World 2020 – Gesture Recognition Demonstrator

Frequently Asked Questions to AIfES:

What devices does AIfES run on?

AIfES^® can be used on almost any system, whether it's a microcontroller, IoT device, Raspberry PI, PC or smartphone. So you do not need to invest in new hardware and instead you can get started right away.

How can I install AIfES?

AIfES^® can be downloaded and installed by searching for "aifes" in the Arduino^® Library Manager. Alternatively, a manual download is also possible. Download the AIfES repository as a ZIP archive and follow the instructions. All information can be found on our GitHub.

For which ML problems is AIfES suitable?

Generally, for any machine learning problem that can be solved with complex neural networks.

Still need inspiration? Our demonstrators can help: gesture recognition, recognition of colors and objects, or an interactive tic-tac-toe-game.

To what extent can the architecture be adapted?

AIfES^® was developed as a flexible and extensible toolbox for the operation and training of artificial neural networks on microcontrollers. All layer, loss and optimization functions are modular and can be optimized for different data types and hardware platforms.Currently complex neural network types for inference and training are supported.

Why is TinyML so exciting?

According to the TinyML Foundation's definition, Tiny Machine Learning (TinyML) is a rapidly growing field of machine learning technologies and applications. It encompasses hardware, algorithms, and software capable of analyzing sensor data on the sensor node at extremely low power (typically in the mW range and below), enabling a variety of applications even on battery-powered devices. Rapid progress is being made. For example, significant advances have been achieved in algorithms, networks, and models with a size of 100 kB and less, and the first low-power applications have been developed in the areas of image processing and audio. AIFES^® is a pioneer and enables data processing on embedded systems. This allows data to be stored on microcontrollers and small IoT devices and processing to be performed without transmission delay. For more on TinyML, see the TinyML Foundation website.