| Afferent input to PNs | Activity of ORNs |

| Activity of neural units driving the latent decoder | Firing rate of PNs with baseline activity set to zero. |

| Activity of auxiliary population of neurons | Firing rate of LNs with baseline activity set to zero |

| Decoded latent space activity: each dimension indicates accrued evidence regardingpresence of high-level stimulus features | Intermediate representation of stimulus information thatdrives downstream behavioral/motor responses |

| Dynamics of the latent space decoder. A represents the internal dynamics of the decoder. Weights of matrix **b** represent combinatorial encoding | |

z | A fixed representation associated with a particular stimulus | |

Q | Penalty on accuracy of latent representation | |

S | Penalty on neural resources (i.e., energy used) | |

R | Regularization on rapid fluctuation in firing rate activity | |

| Connections mapping slow processing of onto its dynamics | |

| Connections mapping fast processing of onto its dynamics | |

| Excitatory connections from the representative PNs to the auxiliary population | Excitatory synaptic connections from PNs to LNs |

| Inhibitory connections from the auxiliary population to the putative PNs | Inhibitory synaptic connections from LNs to PNs |

| Intrapopulation interactions | Synaptic connections among PNs and LNs |

| Computational parameters (matrix) that must satisfy the following: , and | |

C | Fraction of neurons responding to both red/blue stimuli | Percentage of projection neurons that respond to morethan one sensory stimulus |

| Unit variance Gaussian noise for the network model developed through Stochastic Linear Quadratic Regulator formulation | Background noise in neural response |

| Duration of stimulus period | Duration of stimulus period |