Neural model | Description | Similarity measure for two timepoints |
---|---|---|

Posterior | Vector [4 × 1] containing the posterior probability of each sector, P(sector | animals seen so far) | Normalized correlation^{*} |

Posterior–ranking^{**} | Vector [4 × 1] containing the posterior probability of each sector, P(sector animals seen so far) | Spearman correlation |

Log posterior | Vector [4 × 1] containing the natural logarithm of the posterior probability for each sector, log[P(sector animals seen so far)] | Normalized correlation^{*} |

Current animal | An integer ∈{1, 2, 3, 4, 5} indicating which animal is currently on screen | 1 if the same animal; 0 otherwise |

Entropy | A scalar indicating the entropy of the posterior distribution over sectors | −abs[entropy(t_{1}) − entropy(t_{2})] |

MAP | An integer ∈{1, 2, 3, 4} indicating which sector has the highest posterior probability | 1 if the same sector; 0 otherwise |

p(MAP) | A scalar indicating the probability of the MAP sector | −abs[p(MAP(t_{1})) − p(MAP(t_{2}))] |

Posterior–MAPonly | The posterior [4 × 1], zeroed for all sectors except the MAP sector (i.e. a signal that contains both MAP and p(MAP) information) | Normalized correlation^{*} |

Time | A scalar indicating the seconds passed since the start of the session | −abs[time_{1} − time_{2}] |

Posterior–feedbackRL | Vector [4 × 1] indicating the posterior distribution over sectors, computed using the likelihoods updated on each trial using the best-fitting feedbackRL behavioral model (free parameters fitted for each participant) | Normalized correlation^{*} |

MAP–feedbackRL | An integer ∈{1, 2, 3, 4} indicating the most probable sector according to the best-fitting feedbackRL behavioral model (free parameters fitted to each participant) | 1 if the same sector; 0 otherwise |

↵*The normalized correlation of vectors

*x*and*y*is*x*·*y*/(‖*x*‖ ∗ ‖*y*‖), and is equivalent to the cosine of the angle between the two vectors. It behaves differently than the more commonly used Pearson correlation; for example, the posterior distributions [0.24 0.25 0.25 0.26] and [0.26 0.25 0.25 0.24] have Pearson correlation of −1 but normalized correlation of 0.9994. We used normalized correlation because this measure accords better with intuition regarding the similarity of posterior distributions and quantities derived from posterior distributions; however, similar results were observed when using Pearson correlations instead.↵**This neural model is the same as the “posterior” model, except that, by using the Spearman correlation as its similarity metric, it only retains information about the rank ordering of the sectors in the posterior distribution. It is identical to a rank ordering of the sectors in the logposterior model, since the logarithm is a monotonic operator.