| The collective dynamics of neural circuits centrally relies on how individual neurons process their inputs. Despite a vast literature on neural network dynamics, almost all theoretical studies so far have assumed linear summation of inputs. Experimental works, however, have shown that temporally synchronous and spatially close inputs yield a soliton-like excitation and thereby a supralinear enhancement of the inputs. Here we study how such non-additive input processing impacts the dynamics of neural circuits and analytically derive its influence on associative memory networks. Memories are stored into Hopfield-like networks in the form of attractors whose basins reflect the system's error-correcting properties. We show that non-additive coupling effectively extends the network's storage capacity and improves the reliability of memory retrieval against noise. Our results provide a basis for better understanding the impact of more complex connectivity patterns on neural network dynamics in the presence of non-additive coupling. |
|