Treffer: Explainable Artificial Intelligence with MicroPython: Lightweight Neural Networks for Students' Deeper Learning

This study explores the integration of neural networks onto resource-constrained microcontrollers in academic teaching and learning settings in order to enhance students’ understanding of artificial intelligence. By leveraging MicroPython, the open-source framework AI-ANNE: (A) (N)eural (N)et for (E)xploration is used to facilitate the transfer of pre-trained models from high-level artificial intelligence libraries like TensorFlow and Keras to microcontrollers. A comparative analysis was conducted to assess students’ comprehension of neural networks under different instructional methods. Statistical evaluations using Tukey’s HSD post-hoc test revealed that students who implemented neural networks from scratch in MicroPython demonstrated significantly higher learning outcomes compared to those using TensorFlow and Keras in Python. However, no significant difference was observed between students who implemented MicroPython models on standard computing environments and those who deployed them onto microcontrollers. This suggests that explicit implementation in MicroPython fosters a deeper conceptual understanding of neural networks. While high-level artificial intelligence libraries like TensorFlow and Keras provide efficiency, they may obscure fundamental learning processes when it comes to trust, transparency, understandability, and usability as key concepts of explainable artificial intelligence approaches. Thus, direct implementation in MicroPython enhances comprehension and prepares students for responsible artificial intelligence development.