METHOD FOR BUILDING A LARGE GENERAL SPATIO-TEMPORAL AI MODEL

Abstract

The present invention discloses a method for constructing a large general-purpose spatio-temporal AI model, which consists of establishing a topological representation and integrally structured storage of various multimedia information, based on the spatial object representation and the spatial relation model in the form of zero-dimensional objects, one-dimensional objects, two-dimensional objects, and three-dimensional objects; constructing a large spatio-temporal AI model capable of fully representing and processing all objects of human society by means of a training method for a large multimodal model that is represented three-dimensionally; and providing capabilities such as the representation of spatio-temporal relations, semantic understanding, environmental perception, spatio-temporal deduction, and spatial decision-making, so as to solve the problem that the current large language model has only one representation.lack of topological relationships and struggles to address the spatio-temporal relationship of the real three-dimensional world. (Figure for the abbreviation: Fig.1)