Neuromorphic computing architectures
Neuromorphic computing is inspired by biological neural networks and tries to mimic their behavior precisely. Neuromorphic neural networks have advantage over conventional artificial neural networks in their power efficiency and asynchronous, self-adapting nature. Being firstly introduced in the last century, neuromorphic computing is actively rediscovered due to recent advances in the neurobiology, physics, semiconductor technology and AI. But the way neuromorphic architectures work poses new challenges and requires new approaches to the data representation, training, AI-frameworks and tools.
AI in embedded systems
Artificial intelligence is actively adopted in the design of embedded systems. Embedded AI helps localize data flows, needed for training neural networks, what in turn helps to offload the servers, communication network infrastructure and retain the privacy of personal data. With a broad range of existing solutions for AI-accelerated SoCs, it is still challenging to integrate AI into resource constrained systems, due to complexity of neural networks used in industry. Therefore, new optimization techniques on the hardware as well as on the software side have to be researched.
|Deployment of deep neural networks onto embedded devices for processing radar data||Bachelorarbeit|
|Sensor pen trajectory reconstruction using Spiking Neural Networks||Masterarbeit|
Pfau, J.; Leys, R.; Neu, M.; Serdyuk, A.; Peric, I.; Becker, J.
2023. 2023 IEEE International Symposium on Circuits and Systems (ISCAS), 1–5, Institute of Electrical and Electronics Engineers (IEEE). doi:10.1109/ISCAS46773.2023.10181679
Kreß, F.; Serdyuk, A.; Hotfilter, T.; Höfer, J.; Harbaum, T.; Becker, J.; Hamann, T.
2022. 2022 11th Mediterranean Conference on Embedded Computing (MECO). Ed.: IEEE, Institute of Electrical and Electronics Engineers (IEEE). doi:10.1109/MECO55406.2022.9797131