computing system

KAUST-CEMSE-Public-Colloquium-Marco-Canini-Why-ML-Works-Its-Not-Magic

Why ML Works: It’s Not Magic, It’s Networked Systems Wizardry!

-
Today’s machine learning (ML) solutions are achieving extraordinary success across diverse fields, thanks to their ability to learn complex models and deliver impressively accurate results. While the availability of big data is often credited as a driving force, the unsung hero behind ML’s rise is the rapid innovation in computing and software systems that make these breakthroughs possible and accessible.

Low power approaches to communication and computing system design

-

B9 L2 R2322

In this talk we discuss approaches to low power design for advanced communication and computing platforms. Specifically, we present the concept of cognitive power management, where contrary to common approaches that assume a 100% error free hardware, the algorithm is made aware of the statistical error performance of the underlying hardware platform. By accounting for hardware errors at the system level, the explorable power management design space is significantly expanded, leading to novel power saving schemes that deliver expected application performance at much lower power consumption. Sample case studies including LTE system design and in-memory computing platforms will be presented and discussed.