Inhibition enhances capacity of sequence replay: a mean field model

Abstract

Event Abstract Back to Event Inhibition enhances capacity of sequence replay: a mean field model Alvaro Tejero-Cantero1, 2, 3*, Axel Kammerer1, 2, 3 and Christian Leibold2, 3 1 Graduate School of Systemic Neurosciences, Germany 2 Bernstein Center for Computational Neuroscience, Germany 3 LMU Munich, Biologie II, Germany Sequences of neuronal activity patterns can be stored in networks of binary neurons with binary synapses. Patterns are bound into associations via the clipped Hebb rule proposed by Willshaw et al. in 1969 [1]. Such a memory representation is distributed and resilient against local damage. Inspired by hippocampal replay of behavioral sequences [2] we study the capacity limit of a sparsely connected network. Reactivation of pyramidal neurons has been demonstrated to coincide with Sharp-Wave Ripple events (SWR). Recent evidence[4] suggests that each ripple is the framework of expression of a principal cell assembly, and that inhibition follows excitation in time. In order to address the effect of inhibition, we extend on the mean-field model of a CA3-like recurrent excitatory network by Leibold and Kempter [4]. There, it was shown that successful replay requires a minimum sparseness in the code and that the network capacity increases with it. However, stability of replay is lost eventually. Here, we found that the introduction of global inhibition feedback makes sequence replay possible with an sparser code, thereby increasing the memory capacity of the network. At the same time, the range of firing thresholds compatible with replay became broader, suggesting a more robust behavior with noisy, biological neurons. Phase-space analysis with static linear inhibition implemented as a threshold shift showed that inhibition works by adaptively stabilizing unstable replay regimes. The effect of a capacity increase visible under static inhibition was replicated when adding a third, inhibitory population to our mean-field model. The regions of stable replay calculated from both 2D and 3D models were verified in cellular simulations. We conclude that any form of inhibition that offsets the activity background from non memory-participating neurons, whether instantaneous or delayed provides a capacity increase.