Extended Data Fig. 5: Robustness to changes in the single neuron transfer function and recurrent synaptic weights. | Nature Neuroscience

Extended Data Fig. 5: Robustness to changes in the single neuron transfer function and recurrent synaptic weights.

From: Maintaining and updating accurate internal representations of continuous variables with a handful of neurons

Extended Data Fig. 5

Comparison between initial and final bump orientations as a function of JE for a, a network of N = 8 neurons with a Von Mises weight profile and a smooth nonlinear transfer function, and b, a network of N = 16 neurons with a recurrent weight profile storing a 2-dimensional toroidal attractor. In both cases, there is an optimal value of JE for which the circular variance between the initial and final orientations is close to zero (top, red markers), and the bump does not drift (bottom, center panels). Away from these values of JE, the circular variance increases (top, purple/blue markers), and the bump drifts from its original orientation (bottom left/right panels). See Methods for simulation details.

Back to article page