Fecha de publicación: -- Fuente: WIPO "tomato" A method for operating a semantic system having a processor, a memory and a transceiver includes storing a plurality of neural networks, a plurality of capabilities semantics, and a semantic goal. It further includes causing the processor to operate one or more neural network runtimes and to determine a first subset of neural networks based on a semantic projection and the semantic factorization of the plurality of capabilities in rapport with the semantic goal. The processor operates the first subset of neural networks and generates a set of results, and for each result it apply the inferred corresponding factor for the capability semantics associated with the corresponding neural network to determine an aggregate result. The processor then generates semantic augmentation based on the aggregate result.