Luca Pasqualini (DIISM, University of Siena)
Oct 9, 2019 – 11:00 AM
DIISM, Artificial Intelligence laboratory (room 201), Siena SI
Pseudo-Random Numbers Generators (PRNGs) are algorithms produced to generate long sequences of statistically uncorrelated numbers, i.e. pseudo-random numbers (PRNs).
Those numbers are used in mid-level cryptography and in many software applications and libraries.
Test suites are used to evaluate PRNGs quality, the most famous is the one provided by the National Institute of Standard and Techologies (NIST): SP800-22
Machine Learning techniques are often used to break those generators, i.e. approximating a certain generator using a neural network. But what about automatically generating one that is novel from scratch?
What is proposed is a Reinforcement Learning approach to the novel task of generating new, black-box random number generators by maximizing the cumulative expected reward of an N-dimensional pathfinding problem, where N is the length of the sequence to generate and the reward is given by the score of said NIST test suite over that sequence.