Fig. 5.

Memories near attractors are more stable. a If discrete attractors underlie working memory, then memories of different target colors will accumulate error at different rates over time. Memories of target colors near stable fixed points (SFPs) accumulate a relatively small amount of error over time because perturbations away from SFPs due to random noise are corrected by drift back towards the SFP. Unstable fixed points (UFPs) lie in between attractors; perturbations away from UFPs due to random noise are exacerbated by drift away from the UFP. Red dashed lines the indicate location of two of the fixed points. b Mean increase in error at SFPs and UFPs identified by the fit model. Distributions reflect bootstrapped values. *p < 0.05, **p < 0.01, bootstrap. Source data are provided as a Source Data file