Figure 7.

The AD Regime Enables the Precise Encoding of Complex Patterns of Presynaptic Activity, while the RD Regime Exhibits High Population Responsiveness to Afferent Inputs

(A) Representative example of a presynaptic activity pattern that corresponds to 10 activations of different groups of 10 synchronously spiking units (randomly picked within the 100 cells of the presynaptic population) in a 500-ms window (see STAR Methods).

(B) Spiking response of a sub-network of neurons (20 cells) across 20 trials in the AD regime. The y axis indexes both the neuron identity (color coded) and the trial number (vertical extent on a given color level).

(C) Same as in (B) but for the RD regime.

(D) Mean cross correlation of the output spiking patterns across realizations for a given input pattern (mean ± SEM over 10 input patterns; for each input pattern we computed the mean cross correlation across all pairs of observations of the 20 realizations; two-sided Student’s t test).

(E) Performance in decoding the pattern identity from the sub-network spiking patterns with a nearest-neighbor classifier (see STAR Methods). The mean accuracy ± SEM over 10 patterns of 10 test trials each is shown (two-sided Student’s t test). The thin dashed line indicates the level of chance (from 10 patterns and 10 onsets: 1%).

(F) The model network is fed with a stimulus whose firing rate envelope is of varying maximum amplitudes stimulus ${\text{δν}}_{\text{a}}^{\text{max}}$ (amplitude values are color coded).

(G) Mean and standard deviations over n = 10 trials of the increase in excitatory population activity ${\text{δν}}_{\text{e}}\left(\text{t}\right)={\text{ν}}_{\text{e}}\left(\text{t}\right)-{\text{ν}}_{\text{e}}^{\text{stat}}$ in the AD regime. In the inset, the response average in the time window $\text{T}$ (highlighted by a gray bar along the time axis) as a function of the maximum amplitude of the stimulus ${\text{δν}}_{\text{a}}^{\text{max}}$ is shown. Note that the slope in the log-log input-output curves as a function of ${\mathbf{\nu }}_{\mathbf{a}}$ (lower insets in B and C) is not directly informative about the linear gain because of the different ranges in ${\mathbf{\nu }}_{\mathbf{a}}$ and ${\mathbf{\nu }}_{\mathbf{e}}$ (see Figure 4B).

(H) Same as in (F) but for the RD regime.

(I) Slope of the relationship between ${\langle {\text{δν}}_{\text{e}}\rangle }_{\text{T}}$ and ${\text{δν}}_{\text{a}}^{\text{max}}$ (mean ± SEM over n = 10 trials; statistical analysis: two-sided Student’s t test).

(J) Decoding the sub-network rate waveform with a nearest-neighbor classifier. The thin dashed line indicates the level of chance (from the five waveforms shown in F: 20%). The mean accuracy ± SEM over five patterns of 10 test trials each is shown (two-sided Student’s t test).

See also Figure S9.