Speaker
Antal Jakovac
(Wigner RCP, Department of Computational Sciences)
Description
Despite the remarkable achievements of contemporary AI, it is increasingly evident that current architectures suffer from fundamental limitations. One of the most critical weaknesses lies in the representation—indeed, the modelling—of actual reality. A meaningful reality model requires a hierarchical system of partially independent features that together provide a coherent coordination of the environment.
In this talk, I will outline the logic of reality representation and propose principles for constructing effective environmental models. I will discuss the LLT/ALT method for extracting relevant features from continuous data, as well as the use of RDF-style triplet representations for structuring discrete information.
Author
Antal Jakovac
(Wigner RCP, Department of Computational Sciences)
Co-authors
Andras Telcs
Marcell Tamás Kurbucz
(Wigner RCP)
