Webinar organized in collaboration with X-Environnement, the Telecom Paris Alumni Numérique et Environnement group and the Ponts Transition Écologique group.

The need to develop frugal Artificial Intelligence, in other words, sober, is becoming increasingly apparent, due to the very high consumption by AI applications of computing resources, memory resources, and network resources, which translate concretely into the consumption of electronic devices and electricity. Neural networks require ever-increasing resources during their learning phases, and with generative AI, this greed for resources, hitherto limited to learning, has extended to countless user queries. How can we measure and control this explosion? We'll explore a few avenues with our guests:

• Francis Bach, INRIA, Académie des Sciences, will introduce the subject. What do we mean by frugal AI?
• Anne-Laure Ligozat, from the Laboratoire Interdisciplinaire des Sciences du Numérique (LISN), member of the CNRS ECOINFO group, will talk about life-cycle analysis of AI solutions.
• Benoît Courty, from the Codecarbon.io association, will talk about tools for measuring the carbon impact of software.
• Bertrand Charpentier, co-founder of Pruna AI, will talk about reducing the CPU consumption of AI models.