Efficient Inference at the Edge

ABSTRACT: The presentation will review the design space, the trade-offs and the available options for implementing Deep Neural Networks in resource constrained embedded devices. Going into depth, the talk will present:
~ a detailed, layer-wise power and delay profiling of neural network inference, and
~ pruning based network compression techniques.

BIO: Axel Jantsch received the Ph.D. degree from TU Wien, Vienna, Austria, in 1992. From 2002 to 2014, he was a full professor in electronic systems design at KTH. Since 2014, he has been a Professor of systems on chips at TU Wien, Vienna, Austria. His current research interests include systems on chips, self-aware cyber-physical systems, and embedded machine learning. You can read more on his homepage: https://jantsch.se/AxelJantsch/HomePage/research.html

There are a few seats available if you wish to join this seminar in real life on KTH Campus! Please contact vickid @kth.se to register.

TECoSA hosts a guest speaker on the first Thursday of each month during term time. The seminars are free and all are welcome to attend. This Autumn, all seminars will be available via Zoom (with some also IRL at KTH Campus). Please see the TECoSA Seminar Series homepage for details! www.tecosa.center.kth.se/tecosa-seminar-series/