A model of goal-directed spatial navigation based on rodent neurophysiological data

This dissertation presents a biophysical model that combines evidence from neurophysiological studies of goal-directed spatial navigation in the rat into a single algorithm based on bidirectional graph search. Application of this algorithm allows the model to simulate behavioral and neurophysiological data from rats performing memory-dependent spatial tasks. The behavioral component is represented as a virtual animal in a virtual environment. The neurophysiological component is represented as a network of simulated neurons, which receives sensory input from the environment and directs the movement of the animal. The binding of the physiology and behavior in the model allows several experimental phenomena to be addressed, such as (1) activities of the principal cells in the hippocampal area are correlated with an animal's position in space; (2) this positional information degrades after a lesion that disconnects the hippocampus from the medial septum; (3) after lesions of the medial septum, reduction of hippocampal theta rhythm is correlated with an impairment in spatial tasks; (4) cells in the hippocampus tend to fire at a preferred phase of the theta cycle, and (5) the arrival of different inputs to the hippocampus corresponds to specific phases of the theta cycle. The model shows how these physiological properties allow such behavioral functions as successful navigation through various mazes towards a desired goal, selection of a goal on the basis of distance and salience, and reshaping its knowledge about the environment. The latter is based on a thoroughly analyzed spatially and temporally local spike-timing-dependent synaptic modification rule. Four types of gated learning are compared in the context of simulated behavior. Presynaptic and dual OR (triggered by either presynaptic or postsynaptic activity) gatings are more noise tolerant and produce experience-dependent changes in place cell firing patterns shown experimentally. Postsynaptic gating leads to a better competition between input patterns. Dual AND gating (based on the simultaneous presence of both presynaptic and postsynaptic activity) leads to stabilization of the network activity. Each of these gatings can be used to subserve different aspects of the exploratory behavior. The proposed model can be further extended to more general problem-solving based on classical graph search.