Pseudo-random Number Generation Using Binary Recurrent Neural Networks

Loading...
Thumbnail Image
Authors
Hughes, James M.
Issue Date
2007-04-26
Type
Thesis
Language
en_US
Keywords
Research Projects
Organizational Units
Journal Issue
Alternative Title
Abstract
Pseudo-random number generators are deterministic functions that map, in most cases, a state x to a new state x˙ using some update function in order to generate pseudo-random data. Use of these numbers is an integral part of computer science, stochastic physical & statistical simulation, and cryptography. Because of the deterministic nature of these functions, it is impossible to speak of the resultant numbers as truly random. Therefore, the primary goal of pseudo-random number generation is to create values that are statistically identical to truly random numbers. To this end, a few specific characteristics are desirable. First, there should be a way to extract some value from the function (such as a single bit) at discrete intervals that cannot be guessed with probability greater than 50% if only f, the update function, is known. Additionally, functions with long limit cycles and whose cycle lengths grow exponentially in the size of one or more system variables are among those well-suited for random number generation. Finally, the values generated should be independently and identically distributed over the given output interval. Many different, well-established methods exist for generating pseudo-random numbers. Each of these methods has advantages and disadvantages related to its efficiency and effectiveness. In this project, we present the results of a unique pseudorandom number generator created using binary recurrent neural networks trained with two types of neuronal plasticity, anti-spike-timing dependent plasticity (anti- STDP) and intrinsic plasticity (IP). We subject our results to industry-standard random number generator test suites, in addition to performing empirical analysis on the dynamics of our simulated networks. We show that the interaction of these types of plasticity creates network dynamics well-suited for pseudo-random number generation.
Materials made available to the public 3/20/2014 with consent of the author.
Description
39 p.
Citation
Publisher
License
U.S. copyright laws protect this material. Commercial use or distribution of this material is not permitted without prior written consent of the author.
Journal
Volume
Issue
PubMed ID
DOI
ISSN
EISSN