Impact of axonal delay on structure development in a multi-layered network (Q6055153)

In this article, the authors relax \textit{R. Linsker}'s assumption [``From basic network principles to neural architecture: emergence of spatial-opponent cells'', Proc. Natl. Acad. Sci. USA 83, No. 19, 7508--7512 (1986; \url{doi: 10.1073/pnas.83.19.7508}); ``From basic network principles to neural architecture: emergence of orientation-selective cells'', ibid. 83, No. 22, 8779--8783 (1986; \url{doi:10.1073/pnas.83.22.8779}; ``From basic network principles to neural architecture: emergence of orientation columns'', ibid. 83, No. 22, 8779-8783 (1986; \url{doi:10.1073/pnas.83.22.8779})] to examine the impact of distance-dependent axonal propagation delay on neural learning. It is shown that propagation delay induces low-pass filtering by dispersing arrival times of spikes from presynaptic neurons, providing a natural correlation cancellation mechanism for distal connections. Important issues related to expected number of shared inputs, variance of neural activity, covariance of neural activity, covariance of neural activity in the frequency domain, derivation of delay in the frequency domain and derivation of average covariance are discussed in detail in Appendices A)--F).