Summary
Working memories are thought to be held in attractor networks in the brain. Because working memories are often based on uncertain information, memories should ideally come with a representation of this uncertainty for strategic use in behavior. However, the attractor states that hold these memories in attractor networks commonly do not represent such uncertainty. Focusing here on ring attractor networks for encoding head direction, we show that these networks in fact feature all the motifs required to represent uncertainty in head direction estimates. Specifically, they could do so by transiently modulating their overall activity by uncertainty, in line with a circular Kalman filter that performs near-optimal statistical circular estimation. More generally, we show that ring attractors can perform near-optimal Bayesian computation if they can flexibly deviate from their attractor states. Finally, we show that the basic network motifs sufficient for such statistical inference are already known to be present in the brain. Overall, our work demonstrates that ring attractors can in principle implement a dynamic Bayesian inference algorithm in a biologically plausible manner.
Competing Interest Statement
The authors have declared no competing interest.
Footnotes
Changed the focus of the paper to its relevance in working memory; changed title and abstract accordingly; revised results text for clarity; added overview figure; figure 2 (now 3) updated; removed old figure 3; revised methods section for clarity; revised SI for clarity