NEURAL MODEL OF SILENT WORKING MEMORY

Working memory (WM) in cortical systems has been traditionally explained by persistent spiking activity driven by recurrent connectivity [1]. More recently, alternative silent mechanisms have been proposed, in which information is stored in latent neural states and revealed only upon reactivation [2]. Here, we present a mathematically tractable mechanism by which single neurons can silently encode the order of sequentially presented items. The proposed silent WM relies on intrinsic adaptation currents rather than persistent firing. Upon a global activation of the network, differences in adaptation give rise to ordered spike responses that retrieve the stored temporal sequence of inputs. We show that this mechanism produces a spike-time code reflecting the temporal order of presented items, with systematic dependencies on stimulus delay and pulse duration. Finally, we suggest that this principle can be exploited by spiking recurrent neural networks to solve WM tasks. These results highlig​ht intrinsic neuronal adaptation as a viable substrate for silent working memory and temporal sequence encoding.
References
[1] Curtis, Clayton E., and Mark D'Esposito. “Persistent activity in the prefrontal cortex during working memory.” Trends in cognitive sciences vol. 7,9 (2003): 415-423. https://doi.org/10.1016/S1364-6613(03)00197-9
[2] Wolff, M., Jochim, J., Akyürek, E. et al. “Dynamic hidden states underlying working-memory-guided behavior.” Nat Neurosci 20, 864–871 (2017). https://doi.org/10.1038/nn.4546&nbsp