Our societies must face an unprecedented challenge: the climate emergency. We must collectively reflect on a desirable future, respectful of all forms of life on Earth and giving power back to its citizens.
I am a PhD student at Inria de l’Université de Rennes, supervised by Alexandru Costan and Gabriel Antoniu. My research interests cover various aspects of continual learning workloads in distributed settings (HPC, clouds and edge devices). In 2022, I did a student appointment at Argonne National Laboratory, Chicago with Bogdan Nicolae, trying to scale such workloads on many GPUs. My research field could be refered to as “System for AI”.
I hold a degree in electrical engineering from INSA Rennes. In 2019, I did an internship at Snips, Paris working on the integration of an AI voice assistant on edge devices. I then joined Inria as an engineer to develop a software prototype which addresses the limitations of state-of-the-art storage solutions for stream processing.
Check out my profiles at HAL, ORCID, OpenAlex, Scholar.
I participate(d) in the ACROSS project, UNIFY Associate Team, JLESC Towards CL at Scale, PEPR Cloud.
I am a member of Sciences, Environment and Society group in Rennes: we are questioning our research areas in the light of current and future climate challenges. Besides, I have reviewed 10+ papers for IEEE IPDPS, ACM/IEEE CCGrid, ACM/IEEE Supercomputing, IEEE Big Data, ACM HPDC, Springer Computing journal.
I dedicate my skills to initiatives that promote knowledge sharing, the reduction of inequalities and the sustainability of our energy consumption. Finished projects are available here.
Fresks (popularized in France by the Climate Fresk) are serious games where participants explore social or ecological issues in order to understand their stakes. Trouver une Fresque is a search engine that allows you to stay informed about the organization of such workshops in your city.
Intensive code execution generates CO2 emissions: machine learning algorithms typically require the use of GPUs over several days. CodeCarbon is a library allowing to estimate them. The geographical location of the servers is considered in order to take into account the means of electrical production.