Algorithm and Implementation Co-Design for Learning and Inference Slides and Transcript

Abstract- As deep learning comes with significant computational complexity, only relatively recently has this technology become feasible on power-hungry server platforms. In the past few years, we have seen a trend from server-based processing towards embedded processing of the computation for deep learning networks. It is crucial to understand that this evolution is not enabled by either novel architectures or novel deep learning algorithms alone. The breakthroughs clearly come from a close co-optimization between algorithms and implementation architectures. In this presentation, we will review a wide range of recent techniques to a) make the learning algorithms implementation-aware and b) make the hardware implementations algorithm-aware.
Bio- Marian Verhelst is assistant professor at MICAS – KU Leuven, Belgium. Her research focuses on embedded machine learning, low-power sensing, and processing for the internet of things. Marian is member of the Young Academy of Belgium, the ISSCC and DATE executive committees, and is an associate editor of JSSC.