Making Sense Of Raw Input Richard Evans Matko Bošnjak Lars Buesing Kevin Ellis David Pfau Pushmeet Kohli Marek Sergot by Richard Evans & Matko Bošnjak & Lars Buesing & Kevin Ellis & David Pfau & Pushmeet Kohli & Marek Sergot instant download after payment.

How should a machine intelligence perform unsupervised structure discovery over streamsof sensory input? One approach to this problem is to cast it as an apperception task[1]. Here, the task is to construct an explicit interpretable theory that both explains thesensory sequence and also satisfies a set of unity conditions, designed to ensure that theconstituents of the theory are connected in a relational structure.Keywords:However, the original formulation of the apperception task had one fundamental limitation:it assumed the raw sensory input had already been parsed using a set of discreteInterpretable AIcategories, so that all the system had to do was receive this already-digested symbolicUnsupervised theory learningNeuro-symbolic integrationinput, and make sense of it. But what if we don’t have access to pre-parsed input? Whatif our sensory sequence is raw unprocessed information?The central contribution of this paper is a neuro-symbolic framework for distillinginterpretable theories out of streams of raw, unprocessed sensory experience. First, weextend the definition of the apperception task to include ambiguous (but still symbolic)input: sequences of sets of disjunctions. Next, we use a neural network to map raw sensoryinput to disjunctive input. Our binary neural network is encoded as a logic program, so theweights of the network and the rules of the theory can be solved jointly as a single SATproblem. This way, we are able to jointly learn how to perceive (mapping raw sensoryinformation to concepts) and apperceive (combining concepts into declarative rules).© 2021 The Author(s). Published by Elsevier B.V. This is an open access article under the