#EcoLogits featured in the morning presentation at #AIDays in Fribourg:

Why measuring AI environmental impact of organisations is non-trivial?

A paper by Loïc Guibert, David Bekri, Louise Aubet, Steve Berberat, Sébastien Rumley
Link to PDF: https://ai-days.swiss-ai-center.ch/en/papers/block-2

The authors would have liked to use #Apertus instead of Gemma for the LLM-as-judge framework, but cited some incompatibilites at #AIDays that I will look into addressing.

IA4FriLex: Enhancing The Legislative Consultation Process With AI

Ornella Vaccarelli, Beat Wolf, Talia B. Kimber, Thomas Christin, Thomas Kadelbach
https://ai-days.swiss-ai-center.ch/en/papers/block-2

Anastasiia Kucherenko discussing the challenges of traceable LLM Training Data, aka #ExplainableAI

They contributed to the awesome OlmoTrace tool from Ai2, which uses low perplexity scores to match output sequences #AIDays

Leonardo F. Toso takes a break from finishing a Ph.D. to share his research on the intersection of « control theory, machine learning, and optimization, with particular emphasis on meta-learning, federated learning, and adaptive control ».

There is a lot of math in adversarial machine learning, and we are grateful for his masterclass at #AIDays - dig into the award winning papers at https://leotoso.github.io/