La arduino development board It allows thousands and thousands of different projects to be done, the limit is practically in the imagination of each maker, although it also has some physical limitations, such as memory, processing capacity, etc. However, there are more and more products and projects to extend their capabilities more and more, as is the case of the new launch of AIfES.
Now, thanks to this project created by Fraunhofer IMS for Arduino, this open source board will feature a artificial intelligence (AI) framework programmed in C, using the standard GNU GCC compiler libraries. Users will now be able to add AIfES to their Arduino project and integrate it from the library manager from the IDE to use it in your developments, providing the possibility of using machine learning algorithms even in small microcontrollers such as a board Arduino UNO 8-bit.
This will allow developers to create a multitude of IoT (Internet of Things) devices that are more independent from the cloud and that can be more intelligent, and with greater respect for your privacy, since functions can be executed offline from the Arduino board without need to rely on remote services. In addition, the AIfES project has been launched under GNU GPLv3 license, so it is totally free, although it allows a paid license for commercial projects.
AIfES is very similar and compatible with Python ML frameworks as is the case with TensorFlow, Keras or PyTorch, but its functionality is somewhat reduced. However, in this released version the FNN (Feedforward Neural Networks) is already supported, in addition it also allows the activation of integrated functions such as ReLu, Sigmoid, or Softmax. On the other hand, the developers are working to bring in the future also an implementation of ConvNet (Convolutional Neural Networks), which may not take too long to arrive.
Some have also been included training algorithms common, such as SGD (Gradient Descent Optimizer) and Adam Optimizer, among others. I mean, for an 8-bit MCU, it's not bad at all ...