Simple connectionist models of spatial memory in bees

Abstract

Neural network models are used to investigate the ways in which bees use landmarks to navigate through space. The snapshot hypothesis, whereby bees remember a position in space by taking an instantaneous snapshot of the configuration of landmarks, is explored using a Hebbian learning rule and a distributed memory. The number of landmarks and snapshots are shown to contribute to the accuracy of spatial memory. Lining snapshots up along lines of inspection while bees move away from a target site improves performance when estimates of distance involve perceptual errors. When the perception of distance involves scalar error, and if most snapshots are taken close to the target, bees will weight landmarks closer to the target. Networks that respond to distant landmarks can be trained to activate further networks with fine grained local representations to recall the positions of more than one goal. This latter ability compartmentalizes memory for the recollection of complex routes.