Table 2.

Parameter values needed for obtaining numerically similar results to those reported in Cone and Shouval (2021).

Parameter values required for numerical reproducibility
w in	100 nS	Weights of input connections ⊙
σξ	$\mathcal{N}(0,100)$	Gaussian white noise in the neuron model ⊙
d reward	25 ms	Delay of reward signal relative to the onset of the next sequence element ⊘
${\tau }_{\text{syn}}^{\text{exc}}$	80 ms	Excitatory synaptic time constant (EE and IE) within the network ◇
${\tau }_{\text{syn}}^{\text{inh}}$	10 ms	Inhibitory synaptic time constant (EI) ◇
τref	3 ms	Refractory period ◇
φ	0.26	Connection density for all connections (including recurrent) ◇
νin	30 Hz	Rate of the Poisson input ◇
η	0.16	Learning rate for recurrent connections ◇
ηff	20	Learning rate for feedforward connections ◇

Symbols ⊘ and ⊛ as in Table 1. Additionally, ⊙ denotes parameters with no specific values given, while ◇ denotes a mismatch between the values reported in the paper and the ones used in the reference implementation.