Session 1 – 23 March

Towards energy-efficient AI: evolution or revolution?

15:30 – 16:10

Artificial intelligence and machine learning algorithms are slowly becoming mainstream in multiple domains, from industrial diagnostics, healthcare, city management to personal voice assistants and robot vacuum cleaners. As their user base grows, so does the total amount of energy consumed by the creation, training and execution of these algorithms. In particular, the recent introduction of gigantic models, consisting of hundreds of billions of parameters, raised questions on their environmental impact.

In this talk, we place the energy needs of AI in a global context, show how its historical development influenced the energy consumption and point to several hardware and software techniques that improve the situation.  The software approaches are the easiest to apply and reduce the energy consumption without affecting functionality. On the hardware side, a radical departure from conventional digital calculations promises several orders of magnitude improvement. The largest benefit is however found when combining specific software and hardware methods.

Axel Nackaerts studied electronic engineering at the KU Leuven in Belgium, followed by a PhD on physical modeling and digital signal processing. He started at Imec in Leuven in 2003, working on FinFET technology, technology-circuit co-design and design for manufacturing. In 2007, he moved to NXP Semiconductors Corporate Research. He became a system architect for healthcare products and innovation manager, responsible for multiple national and European funded projects. In 2020, he took the role of program manager artificial intelligence at Imec and is the Challenge Manager for “Energy-efficient AI at the edge,” one of the four Grand Challenges of the Flanders AI program.