Path Integration and Cognitive Mapping in a Continuous Attractor Neural Network Model

摘要: A minimal synaptic architecture is proposed for how the brain might perform path integration by computing next internal representation of self-location from current and perceived velocity motion. In model, a place-cell assembly called “chart” contains two-dimensional attractor set an “attractor map” that can be used to represent coordinates in any arbitrary environment, once associative binding has occurred between chart locations sensory inputs. hippocampus, there are different spatial relations among place fields environments behavioral contexts. Thus, same units may participate many charts, it shown number uncorrelated charts encoded recurrent network potentially quite large. According this theory, firing given cell primarily cooperative effect activity its neighbors on currently active chart. Therefore, not particularly useful think cells as encoding particular external object or event. Because connections, hippocampal field CA3 possible location “multichart” architecture; however, other implementations anatomy would invalidate main concepts. The model implemented numerically both integrate-and-fire “macroscopic” (with respect space states) description system, based continuous approximation defined system stochastic differential equations. It provides explanation hitherto perplexing observations fields, including doubling, vanishing, reshaping distorted environments, acquiring directionality two-goal shuttling task, rapid formation novel slow rotation after disorientation. makes several new predictions about expected properties network.