A walk between Topos theory and Artificial Intelligence: examples, interactions, hints for possible future explorations.
报告人：Aurélien Sagnier（华为拉格朗日数学与计算研究中心）
时间：2025-12-11 15:30-16:30
地点：智华楼四元厅
报告摘要 : Defaults of current Artificial Intelligences are the lack of explainability, the lack of certifiability and the huge cost of training (both in computing power and in size of training datasets). In this lecture, I will try to present some material to convince the audience that Topos theory has the necessary tools to overcome those current limitations.
The following questions will be the red threads of this lecture :
-how some limitations of current Artificial Intelligences call for the use of topossic inspirations or technics?
-how topos theory could offer notions to encode and handle data in a more meaningful, explainable and reliable way in Artificial Intelligence?
I will expose during this lecture materials from the state of the art as well as from some preliminary works of my former team in Huawei AWT Paris and of my current team in Huawei’s Paris Lagrange Center. The choice of exposed material is personal and thus non exhaustive. My choices were heavily influenced by the deep theoretical insights of Professor Olivia Caramello, Professor Laurent Lafforgue and by the practical insights of my teams in Huawei Paris.
报告人介绍：Aurélien Sagnier is a mathematician who obtained his PhD in 2017. His first field of study was topos theory and number theory (especially the program of Connes and Consani in number theory). He has joined Huawei in August 2022 as a research engineer. He is now in charge of an internal project whose aim to apply to AI the topossic insights of Professsor Olivia Caramello and of Professsor Laurent Lafforgue."
