How the (Tensor-) Brain uses Embeddings and Embodiment to Encode Senses and Symbols
arXiv Preprint Archive September 19, 2024 Volker Tresp, Hang Li
The Tensor Brain (TB) is a computational model of perception and memory with two layers: a representation layer modeling the subsymbolic global workspace and an index layer containing symbolic labels for concepts, time, and predicates. Sensory input activates the representation layer, which triggers associated symbols (bottom-up), while symbols can also activate the representation layer to influence perception (top-down), enabling semantic memory. Concept embeddings serve as connection weights linking the layers, consolidating knowledge from diverse experiences and modalities into a unified learning and memory framework.