BSidesSF 2018

Wouldn't it be awesome to build a machine learning device that ran on tubes, valves, and gears?

Terms like machine learning, deep learning, and neural nets are often brought up as if they are a magical cure for security problems. Unfortunately, machine learning systems have fundamental, inescapable limitations. Exploring the limitations is normally done through discussion of the mathematics involved. Instead of using math, we will explore the limitations using a steampunk model.

In this presentation we cover the essential elements of neural nets used for machine learning. Instead of using math, we will go over how to build a physical implementation of a machine learning system using tubes, valves, and gears. With the model we will then explore how and why the machine generates false positives.