Table 3.

ARTFLOW hyperparameters and default values used in simulations. Most MT hyperparameters are set according to physiologically informed values.

Parameter	Layer Type	Description	Value
$N_{MT}$	MT	Number of MT neurons	5000
	MT	RF radius	15 pixels (2.5°)
${\theta }_{pref}$	MT	Preferred direction of neuron	Uniform (0°, 360°)
${\sigma }_{\theta }$	MT	Bandwidth of direction tuning	3°
${\sigma }_{\nu }$	MT	speed tuning bandwidth	$\mathcal{N}\left({\mu }_s=1.16°,{\sigma }_s=0.5°\right)$
$s_0$	MT	Non-negative offset parameter to prevent singularity in speed logarithm	Exp $\left(0.25°/\mathrm{s}\right)$
${\nu }_{pref}$	MT	Preferred speed of neuron	Uniform $\left(0.5,2.0\right)°/\mathrm{s}$, Uniform $\left(2.0,4.3\right)°/\mathrm{s}$, Uniform $\left(4.3,7.6\right)°/\mathrm{s}$, Uniform $\left(7.6,12.7\right)°/\mathrm{s}$, Uniform $\left(12.7,32.0\right)°/\mathrm{s}$
$A_{MT}$	MT	Passive decay rate	0.1
$B_{MT}$	MT	Excitatory upper bound	2.5
	MT	Temporal integration time step	0.1 frames (0.003 s)
${\gamma }_{MT}$	MT	Input value that yields 0.5 in sigmoid activation function	0.007 (median empirical input)
	Fuzzy ART	Fuzzy ART module arrangement	Layer 1: 8 × 8, layer 2: 1 × 1
$\alpha$	Fuzzy ART	Activation prior on existing representations	0.01
$\rho$	Fuzzy ART	Vigilance	Layer 1: 0.65, layer 2: 0.85
$\beta$	Fuzzy ART	Learning rate	1 (when committing new cell)0.1 (when updating weights)
	Fuzzy ART	Number of training epochs	Layer 1: 1, layer 2: 1