Training: 8–9 November 2016
Tutorials & Conference: 9–11 November 2016
Amsterdam, NL

Machine learning to improve random number generators

Richard Freytag (Freytag & Company, LLC)
15:50–16:30 Friday, 11 November, 2016
Tech, tools, and processes
Location: G104/105 Level: Intermediate
Average rating: *****
(5.00, 1 rating)

Prerequisite knowledge

* How to apply machine learning to test pseudo-random number generators. * A quick and practical introduction to AzureML Studio. * How to interpret machine learning results in AzureML. * Explore some current limitations of machine learning

What you'll learn

  • Understand how to apply machine learning as a black box in pseudo-random number generators


Tests of pseudo-random number generator (PRNG) performance use deterministic analysis to expose weaknesses, which new PRNGs are designed to satisfy. Modern supervised learning algorithms offer an improved method to test PRNG performance. Richard Freytag offers a short, concrete, and intuitive exploration of how to apply machine learning as a black box in pseudo-random number generators.

Topics include:

  • How we test pseudo-random number generators (PRNGs) now
  • John von Neumann’s first PRNG
  • A slightly better PRNG: The linear congruential generator
  • MD5
  • Applying machine learning to PRNGs
  • How a new PRNG can be developed with machine learning
  • How a new, machine-learned PRNG can help in finding weaknesses or backdoors in future new generators
Photo of Richard Freytag

Richard Freytag

Freytag & Company, LLC

Richard Freytag is the owner of a small development and contracting company, Freytag & Company, whose clients have included the Department of Defense. Most of Freytag & Company’s products target the .NET platform and range from email add-ins, phone apps, and desktop applications to SaaS offerings in the cloud, and its most important products combine machine learning and computer security as core features.