Edge-AI in Software and Sensor Applications
- AI in your Pocket -
Instructor: PD Dr. Victor Pankratius
Winter Semester 2024-2025
All digital course offered through KIT, Fridays 08:00 - 09:30
First lecture: October 25, 2024
Registration required in CAS system (course #2400006)
COURSE REACHED FULL CAPACITY, registration closed
Registered students will be contacted before the first lecture with content access instructions.
For questions or in case of problems with your registration, please contact: firstname.lastname at kit dot edu
Description
Just imagine a world, where every thing you touch is intelligent and ready to assist you. Where everyday devices learn with you autonomously all the time, augmenting your senses and providing immediate feedback for your decisions. EdgeAI is the next frontier in artificial intelligence that enables such capabilities in the smallest imaginable devices and sensors even when there is no cloud connectivity.
Edge Computing includes applications, data, services at the periphery of networks that are close to real-world sensors. Edge systems are typically constrained in their available energy budget, CPUs, memory, and connectivity. Fog computing further combines these aspects with cloud architectures in order to add enhanced local pre-processing and intelligence that extends the capabilities of classical clouds.
Modern sensor applications - for instance in industrial monitoring and logistics, Internet-of-Things, Ubiquitous Computing, mobile devices, wearables & hearables, health & fitness, drones, or augmented reality - increasingly rely on Edge and Fog Computing to better handle Big Data, always-on applications, continuous fusion of data streams, and new kinds of use cases that were unimaginable before.
In this context, Edge Artificial Intelligence methods (Edge-AI) become key to the realization of continuously learning systems that provide more autonomy and instant feedback. In contrast to mainstream AI, EdgeAI techniques have to cope with significant resource constraints and be fault-tolerant. This course therefore picks up on this exciting topic to provide an overview of state-of-the-art, further dive into current research works, show demonstrations, and discuss open problems.
Suggested Literature
Fog and Edge Computing: Principles and Paradigms, R. Buyya & S. N.Srirama, Wiley 2019, ISBN 978-1119524984
TinyML: Machine Learning with TensorFlow Lite on Arduino and Ultra-Low-Power Microcontrollers, P. Warden & D. Situnayake, O'Reilly 2019, ISBN 978-1492052043
Edge-Oriented Computing Paradigms: A Survey on Architecture Design and System Management, Li et.al., ACM Computing Surveys 51(2), 4/2018, https://doi.org/10.1145/3154815
Practical Deep Learning for Cloud, Mobile & Edge, A. Koul et.al., O'Reilly, 10/2019, ISBN 978-1-492-03486-5
Machine Learning for Data Streams, A. Bifet et.al., The MIT Press, 2017, ISBN 978-0-262-03779-2
Artificial Intelligence: A Modern Approach, S. Russel & P. Norvig, 2022, ISBN 978-1-292-40113-3
Fundamentals of Machine Learning for Predictive Data Analytics, Kelleher et al., The MIT Press 2020, ISBN 978-0262044691
Machine Learning: A Probabilistic Perspective, K.P. Murphy, The MIT Press 2013, ISBN 9780262018029
Handbook of Modern Sensors, J. Fraden, Springer 2016, ISBN