Fig. 3
Noise amplified by chaos determines divergence. a1 Time-course of correlation rV after resuming at t0 from identical conditions with different forms of perturbation. Full cellular noise as before, solid line (abcd); no cellular noise, but perturbing with a single extra spike in one neuron, dashed line (f); a miniscule step-pulse perturbation in all neurons, dotted line (e). (abcd: 40 saved base states; e, f: 20 saved base states; mean ± 95% confidence interval). a2 Steady-state root-mean square deviation RMSD∞ and correlation r∞ for stochastic (abcd) and deterministic simulations (e, f) as defined in a1 (mean ± 95% confidence interval in black; individual base states in purple dots). b1 As in a2, but for decoupled, replayed simulations. (20 saved base states). b2 Similarity sRMSD and sr at 10–20 ms with all noise sources enabled, for network and decoupled simulations (mean ± 95% confidence interval). c1 Decoupled replay paradigm. Presynaptic spike trains from a network simulation are saved and then replayed to the synapses of each neuron in a decoupled simulation, thereby removing variability due to feedback network dynamics. c2 Overview of sources of noise and perturbations. d Decoupled replay simulations (see c1) for a representative L4 PC neuron, with somatic membrane potential differences between the two trials only due to cellular noise sources (ab[c]d), a single extra presynaptic spike (f) or a miniscule step-pulse perturbation (e). [c] indicates that for some neuron types in the NMC-model, such as L4 PCs, no stochastic ion channels are present
