Figure 3.

Attractor network models of decision-making. (a) The full model (top) contains one general (NS) and two input-dependent excitatory populations (1 and 2) and one inhibitory population (I) of neurons. This complex model is well approximated by a two-population model (bottom) that only tracks the average activity of the two input-dependent excitatory populations [49]. (b) Schematic illustration of the energy landscape of the reduced model for balanced inputs. Initially, evidence is accumulated in the flat area of the energy landscape towards the unstable saddle point (blue empty dot), until the network state ‘drops’ into either of the two stable basins of attraction (blue solid dots), at which point a decision is made. The red and green traces provide two examples of network state trajectories, each leading to a different decision. An imbalanced input reshapes the energy landscape to make it more likely for the network state to reach the point attractor associated with the stronger input [49].