A neuro-vector-symbolic architecture for solving Raven’s progressive matrices

Neither deep neural networks nor symbolic artificial intelligence (AI) alone has approached the kind of expressed in humans. This is mainly because are not able to decompose joint representations obtain distinct objects (the so-called binding problem), while AI suffers from exhaustive rule searches, among other problems. These two problems still pronounced neuro-symbolic AI, which aims combine best paradigms. Here we show that can be addressed with our proposed neuro-vector-symbolic architecture (NVSA) by exploiting its powerful operators on high-dimensional distributed serve as a common language between and AI. The efficacy NVSA demonstrated solving Raven’s progressive matrices datasets. Compared state-of-the-art network approaches, end-to-end training achieves new record 87.7% average accuracy RAVEN, 88.1% I-RAVEN Moreover, compared reasoning within probabilistic less expensive operations orders magnitude faster. Neuro-symbolic approaches display both perception capabilities, but inherit limitations their individual learning components. By combining vector-symbolic architectures, Hersche colleagues propose framework solve tests faster more accurately than methods.