Applications of artificial intelligence (AI), such as those driven by deep learning, are having great impact on numerous sectors of societal importance, such as healthcare, transportation, agriculture, and computer security. An important trend in deployment of AI applications and systems is the migration of signal and information processing functionality closer to the point at which the data is generated or captured — that is, migration toward the network edge as opposed to the performing all of computation on centralized cloud servers. Edge processing offers various potential advantages, including the potential to greatly reduce delays associated with network communication, enhance privacy, and improve reliability and predictability in scenarios where network performance exhibits significant variation.
This summer school will cover key concepts and methods in design and implementation of edge processing systems for AI applications. The summer school will involve lectures and hands-on laboratory sessions on topics that include parallel programming for embedded multiprocessor systems-on-chip (MPSoCs); technology, architecture and organization of memories used at the edge; hardware accelerators for edge processing; real-time scheduling analysis; hardware-friendly reinforcement learning; and Markov decision processes for power/performance optimization.
The summer school will include a keynote lecture by Prof. Marilyn Wolf, Elmer E. Koch Professor of Engineering and Director of the School of Computing at the University of Nebraska-Lincoln.
The summer school is targeted primarily to Ph.D. students and early-stage post-doctoral researchers. Participants will have opportunities to present their research, including both work-in-progress or published research, in poster sessions at the event.
The lectures and labs will be in English. Non French speakers are welcome.
- Clic for a better resolution :
- Keynote lecture by Prof. Marilyn Wolf
- RTS : Real-time scheduling analysis of SDF graphs: an example with Cheddar (Hai Nam Tran, Lab-STICC)
- TPG : Hardware Friendly Reinforcement Learning with Tangled Program Graphs (Karol Desnos, IETR)
- PP : Parallel Programming for Embedded MPSoCs (Maxime Pelcat, IETR)
- OCM : On-chip memories at the edge: the edge of memory (Kevin Martin, Lab-STICC)
- DSE : Design Space Exploration for machine learning hardware accelerators (Steven Derrien, IRISA & Mickael Dardaillon, IETR)
- MDP : Markov Decision Processes for Power/Performance Optimization (Shuvra S. Bhattacharyya, UMD)
If you are interested in registering for the summer school, please contact Jean-François NEZAN. Space is limited to 20 students so it is suggested that interested students get in contact soon. There is no registration fee and moreover, accommodation during the period June 13-16 is covered by the summer school. To facilitate travel to and from the event, the first and last day are planned as half-days, while the middle day is a full-day.
Accomodation : smart-appart.fr
Kevin Martin, Lab-STICC
Shuvra S. Bhattacharyya, INSA/IETR and UMD
The summer school is organized by INSA/IETR-VAADER, INRIA-TARAN, INRIA-TEA, Lab-STICC-ARCAD, and LabSTICC-SHAKER, and is sponsored by the CominLabs International Chair on Adaptive Machine Learning at the Network Edge and by UBS.