Figure 5.

Computing a linearly non-separable function (full FBP) with supralinear and sublinear dendrites and using local vs. global synaptic wiring strategies. (A) Left, model neuron with equivalent dendrite representation of two compartments, linear (black) and supralinear (green), and a clustered distribution of object features (object 1 : ×1, ×2 and object 2 : ×3, x4) (local strategy). Right, schematic representations of synaptic placements equivalent to the model on the left. (B) Left, model neuron with equivalent dendrite representation of two compartments, linear (black) and sublinear (blue), and a distributed placement of inputs carrying object features. (C–E) Implementation of the full FBP (y = 1; “apple shape and red” or “banana shape and yellow”). (C), implementation of the full FBP using a model with a supralinear compartment and a local wiring strategy. Inactive inputs are represented in light gray and the corresponding feature in lighter color. (D) Implementation of the full FBP using a model with a supralinear compartment and a global wiring strategy. The area of the disc adjacent to a compartment next to each object feature represents the relative weight of this feature. Here the relative weights used are of 1 and 2. (E) Implementation of the full FBP using a model with sublinear compartment and a global wiring strategy.