Table 1.

Best fitting parameter values and fit statistics (χ2 and A/C) for all model architectures and datasets

Dataset/architectureχ2AICNθkgβ7.7β14β19β20u7.7u14u19u20
Pooled
    Gated race62365831301.3370.4300.427
    Gated diffusion-like2656398402.6580.0010.5110.0390.0500.0410.050
    Gated competitive123619910811.6050.0170.3300.0480.0380.0240.024
    Nongated, nonleaky234658355060.3020.0030.0040.0030.002
    Nongated, leaky16562378017.7070.0220.0040.0220.0060.013
Monkey Q
    Gated race4036543663.3730.0180.599
    Gated diffusion-like2446624663.2980.0010.5200.0480.0280.0410.047
    Gated competitive106649713413.6070.0140.3300.0500.0340.0180.021
    Nongated, nonleaky331691745060.2380.0040.0040.0040.004
    Nongated, leaky159643410017.5730.0270.0310.0270.0140.012
Monkey S
    Gated race6316193501.0950.5190.500
    Gated diffusion-like3516120451.0010.4240.3770.0400.0500.0070.050
    Gated competitive1576006709.8160.0280.2690.0040.0470.0000.025
    Nongated, nonleaky324641150055.3630.0030.0050.0030.004
    Nongated, leaky18160217014.0180.0270.0080.0300.0080.020
  • N: number of spike trains sampled from visually-responsive neurons to generate salience input for all conditions. θ, threshold; k, leakage constant; g, gate constant; βd, lateral inhibition weights between two units spaced i degrees of visual angle apart; ud, feed-forward inhibition weights between two units spaced i degrees of visual angle apart. Bold values indicate the best fitting model (minimum χ2). Dashes indicate the parameters were fixed to zero for a given architecture.