Machine Learning on Microcontrollers

Course Description

Contents

Artificial Intelligence (AI) is transforming the digital world, and today even low-power embedded systems are becoming smart and adaptive. This course introduces Machine Learning (ML) for microcontrollers and edge devices, showing how intelligence can be deployed directly at the source of data — sensors and IoT nodes.

Students will learn both the theoretical foundations and practical techniques required to design and implement ML algorithms on resource-constrained hardware. The course covers data acquisition, feature extraction, efficient ML models, and energy-optimized deployment on real devices such as ARM Cortex-M microcontrollers and Edge AI accelerators (MAX7800, GAP9, IMX500).

Alongside the lectures, hands-on laboratory sessions allow students to experiment with real hardware, collect sensor data, train and compress models, and deploy them on microcontrollers. Students gain experience across the full edge AI pipeline — from training to real-time inference.

Learning Goals

• Gain an in-depth understanding of machine learning principles tailored for low-power embedded systems.
• Explore embedded system architectures and learn how to efficiently run ML algorithms on microcontrollers.
• Study classical ML algorithms (Decision Trees, Random Forests, kNN, SVMs) and deep learning models (CNNs, Transformers) for edge deployment.
• Learn optimization techniques such as quantization, pruning, and knowledge distillation to meet memory and latency constraints.
• Understand hardware-software co-design for efficient embedded ML implementations.
• Get familiar with modern Edge AI platforms like ARM Cortex-M, MAX7800, GAP9, and IMX500.
• Acquire hands-on skills in sensing, data collection, and embedded programming using C/C++ and ML frameworks.
• Evaluate performance, accuracy, and power efficiency of ML models running on microcontrollers.

Platforms for MLoM

The practical sessions complement the lecture by allowing students to design, program, and test ML-enabled embedded systems using real hardware and datasets.

The main lab platforms include the STM32U5 and MAX7800, while advanced students can experiment with GAP9, IMX500, STM32N6, Renesas M85, Ambiq, or Coral Dev Micro for higher performance and grading potential.

Hands-on tasks include:

• Image and Sound Classification on microcontrollers
• Model Optimization through quantization, distillation and pruning
• Energy efficiency profiling and benchmarking