Noisy recurrent neural networks: the discrete-time case

O Olurotimi; S Das

doi:10.1109/72.712165

Noisy recurrent neural networks: the discrete-time case

IEEE Trans Neural Netw. 1998;9(5):937-46. doi: 10.1109/72.712165.

Authors

O Olurotimi¹, S Das

Affiliation

¹ Department of Electrical and Computer Engineering, MS 1G5, George Mason University, Fairfax, VA 22030, USA.

PMID: 18255777
DOI: 10.1109/72.712165

Abstract

The results of the companion paper on the behavior of continuous-time recurrent neural networks (RNN) are specialized to the discrete-time (so-called time-lagged recurrent) case. Uniform boundedness of the first and second moment sequences of the trajectory is established. For practical design purposes, estimates and bounds on the bias and variance sequences of the stochastic discrete-time RNN are derived. A script file for estimating the variance bounds for specified designs is provided in an Appendix.