Can Dynamic Neural Filters Produce Pseudo-Random Sequences?

Abstract. Dynamic neural filters (DNFs) are recurrent networks of binary neurons. Under proper conditions of their synaptic matrix they are known to generate exponentially large cycles. We show that choosing the synaptic matrix to be a random orthogonal one, the average cycle length becomes close to that of a random map. We then proceed to investigate the inversion problem and argue that such a DNF could be used to construct a pseudo-random generator. Subjecting this generator’s output to a battery of tests we demonstrate that the sequences it generates may indeed be regarded as pseudo-random.

• Type: Conference
• Year: 2005
• Where: ICANN
• Authors: Yishai M. Elyada, David Horn
• Keywords: Average Cycle Length, Dynamic Neural filters, ICANN 2005, Synaptic Matrix