A computational model of the neural circuits subserving spatial memory and navigation.

摘要

Evidence from both behavioural and electrophysiological experiments in animals and humans demonstrates the existence of world-centered (allocentric) and viewer-centered (egocentric) representations of space in the brain. Additionally, allocentric representations tend to be associated with the medial temporal lobes and, hence, long-term memory, while egocentric representations tend to be associated with posterior parietal cortex and are often of a more transient nature. How these different types of spatial representation interact to allow the brain to generate adaptive behaviour is not well-understood. For the case of long-term spatial memory, we argue that the layouts of familiar environments are encoded as allocentric representations in the medial temporal lobes and that these stored representations are only accessible via a limited egocentric "window" supported by neural circuitry of the posterior parietal cortex. We support this claim by constructing a neural network model of medial temporal-posterior parietal circuitry based upon this principle. The model, which is constrained by neuroanatomical and electrophysiological data, is able to explain experimental findings at both the behavioural and neural levels, including the ability to recall spatial layouts and the ability to engage in spatial updating or mental exploration based upon recalled representations. A hand set neural circuit which transforms between egocentric and allocentric representations forms the core of this long-term memory model. However, it is assumed that a biological counterpart to this component would have to acquire its transformation ability via learning. We present principles that might underlie such a learning process, along with modeling work which demonstrates that they can lead naturally to the learning of place cell responses under certain circumstances. Furthermore, the nature of interaction between long and short-term spatial memory in situations where familiar and novel spatial information must be used in concert to accomplish a task is addressed.