Tasks
1.Summarize i) key terms and concepts in the field of Machine Learning (ML) and ii) ML instruments applicable in the area of embedded systems; especially, focus yourself on microcontroller (MCU) based ML solutions (e.g., TinyML, TensorFlow Lite Micro, OONX, CMSIS-NN, ARM NN, eIQ, Edge Impulse, MATLAB and Simulink for Edge AI) in the context of resource-limited platforms such as Arduino, ESP32, or ARM Cortex-M. 2. Do a literature review in the field of ML solutions as well as in the field of their applications, platforms used, results they achieved etc. 3. Choose an embedded, MCU based, system, a representative (your own, an existing etc.) problem solvable by ML on the system and ML means to implement your solution. 4. Implement a solution from the item 3, check its basic usability and properties. 5. Make a set of experiments to show how attributes (e.g., memory footprint, latency) of the solution depend on the target platform, configuration etc.
Requirements
Any previous experience with machine learning and embedded systems is welcome. Active interest in the topic, creativity, ability to solve problems independently as well as ability to (self) study are strongly recommended.
Expected Outcomes
Representative ML model(s) and their evaluation in the context of their deployment on embedded platforms, a short (about 10 pages long) technical report.
References
According to the promoter’s/supervisor\\\'s recommendation – e.g., you can start here:
https://edgeimpulse.com/,
https://paperswithcode.com/task/anomaly-detection,
https://paperswithcode.com/dataset/toyadmos,
https://backend.orbit.dtu.dk/ws/portalfiles/portal/274322308/ A_Primer_for_tinyML_Predictive_Maintenance_Input_and_Model_Optimisation.pdf.