An Analysis of a Ring Attractor Model for Cue Integration

Animals and robots must constantly combine multiple streams of noisy information from their senses to guide their actions. Recently, it has been proposed that animals may combine cues optimally using a ring attractor neural network architecture inspired by the head direction system of rats augmented with a dynamic re-weighting mechanism. In this work we report that an older and simpler ring attractor network architecture, requiring no re-weighting property combines cues according to their certainty for moderate cue conflicts but converges on the most certain cue for larger conflicts. These results are consistent with observations in animal experiments that show sub-optimal cue integration and switching from cue integration to cue selection strategies. This work therefore demonstrates an alternative architecture for those seeking neural correlates of sensory integration in animals. In addition, performance is shown robust to noise and miniaturization and thus provides an efficient solution for artificial systems.