Sensory Representation Spaces In Neuroscience And Computation

摘要

Physics, Neuroscience and Computation are concerned with finding the most appropriate representation spaces to describe the interaction of a dynamic system with its environment. In this work first we review the two basic conceptual approaches to the problem of representing an environment, Marr's ascending "constructivism" and Gibson's "direct perception" hypothesis. Later we review the basic neural mechanisms associated with creating meaning in both approaches: lateral inhibition and the creation of cortical maps by resonance to patterns of stimuli of families of spatially ordered neurons. We end by considering the usefulness in artificial intelligence of knowledge about the way in which biological systems construct their representation spaces. We stress the idea regarding events as representation entities and, consequently, using an event time, different from physical time. Semantics emerges from the mechanisms that detect these relevant events in each organisational level and their composition rules to specify the constitutive entities of the next level. This semantic is distributed in the cortical maps of the neuron groups that resound to the corresponding events.