时间：2026年6月16日（星期二）16:00-17:00
地点：西湖大学云谷校区E13-105
主讲人：Olivia Caramello, University of Insubria, Italy and Grothendieck Institute
报告题目：Toposes and 'bridges' for artificial general intelligence
报告摘要：Grothendieck toposes, geometric logic and relative toposes offer a foundational framework for artificial general intelligence by modelling information as hierarchical semantic structures rather than purely numerical data.
Through classifying toposes and topos-theoretic 'bridges', this approach extracts invariant properties across diverse representations and mirrors human cognition by organizing knowledge into stratified, explainable levels of abstraction.
When applied to tasks like Raven's matrices or ARC-type problems, the framework drastically reduces parameter search spaces, enables formally verifiable logic, and offers a pathway to integrate structured mathematical reasoning with contemporary deep learning and large language models.
