Embedded AI (EAI)

This is the course “Embedded AI” or short EAI at TH Rosenheim. The lecture is given in the summer semester for the bachelor program Applied Artificial Intelligence.

The responsible person for the module is Prof. Dr.-Ing. Kevin Mayer.

In the first part of the module, the challenges of bringing AI to the edge and the limitations of embedded systems are identified. Algorithms are analyzed with regard to computational complexity, memory footprint, and timing. As of today, typical AI applications are distributed across edge, edge servers, and the cloud. Distributed architectures are discussed, and their advantages and disadvantages are illustrated with examples. We will also view secure and robust models in the context of cybersecurity.

The second part of the lecture focuses on hardware acceleration of AI applications on embedded devices. Different classes of embedded hardware and accelerators (TPU, DSP, FPGA, …) are introduced and analyzed, and their limiting factors are identified. Here, we will conclude the section on hardware security, which is highly relevant to embedded devices.

In the third part of the lecture the students will learn about algorithmic as well as software optimizations that are typically applied to further speed up computations on embedded systems. We will also touch the basics of secure AI lifecycles. The fourth of the lecture focuses on threat model in embedded AI, attacks and defenses on edge AI, and common mistakes made in embedded AI. In addition to the lecture there will be practical exercises and lab courses allowing students to bring AI algorithms to life on different embedded systems and analyze the results.

The course can be found on Learning Campus.