Interactions of spatial strategies producing generalization gradient and blocking: A computational approach

We present a computational model of spatial navigation comprising different learning mechanisms in mammals, i.e., associative, cognitive mapping and parallel systems. This model is able to reproduce a large number of experimental results in different variants of the Morris water maze task, including standard associative phenomena (spatial generalization gradient and blocking), as well as navigation based on cognitive mapping. Furthermore, we show that competitive and cooperative patterns between different navigation strategies in the model allow to explain previous apparently contradictory results supporting either associative or cognitive mechanisms for spatial learning. The key computational mechanism to reconcile experimental results showing different influences of distal and proximal cues on the behavior, different learning times, and different abilities of individuals to alternatively perform spatial and response strategies, relies in the dynamic coordination of navigation strategies, whose performance is evaluated online with a common currency through a modular approach. We provide a set of concrete experimental predictions to further test the computational model. Overall, this computational work sheds new light on inter-individual differences in navigation learning, and provides a formal and mechanistic approach to test various theories of spatial cognition in mammals.Author summary: We present a computational model of navigation that successfully reproduces a set of different experiments involving cognitive mapping and associative phenomena during spatial learning. The key ingredients of the model that are responsible for this achievement are (i) the coordination of different navigation strategies modeled with different types of learning, namely model-based and model-free reinforcement learning, and (ii) the fact that this coordination is adaptive in the sense that the model autonomously finds in each experimental context a suitable way to dynamically activate one strategy after the other in order to best capture experimentally observed animal behavior. We show that the model can reproduce animal performance in a series of classical tasks such as the Morris water maze, both with and without proximal cues, which support the cognitive mapping theory. Moreover, we show that associative phenomena such as generalization gradient and blocking observed within the navigation paradigm cannot be explained by each learning system alone, but rather by their interaction through the proposed coordination mechanism. The fact that these experimental results have for a long time been considered contradictory while they could here be accounted for by a unified modular principle for strategy coordination opens a promising line of research. We also derive model predictions that could be used to design new experimental protocols and assess new hypotheses about complex behavior arising from the interaction of different navigation strategies.