Hacking Machine Learning Algorithms

Presented at The Eleventh HOPE (2016), July 22, 2016, 10 p.m. (60 minutes).

Algorithms control more and more of the systems we interact with on a daily basis. Critical decisions are executing without direct oversight by machine learning models. These systems, like any system, should be continuously taken apart and inspected to see how they work. Examining a machine learning model is not as easy as examining source code. This talk goes into detail on how to hack machine learning models and similar systems. Could an algorithm be racist? How can we detect it? Live examples in Python will be demoed and available on GitHub, and only basic programming knowledge is required to understand the talk and reproduce the examples.

Presenters:

• Kyle Polich
  Kyle Polich has been a data scientist for over a decade. He hosts the Data Skeptic podcast that explores the intersection of data, machine learning, statistics, and scientific skepticism.

Links:

• https://xi.hope.net/schedule.html#-hacking-machine-learning-algorithms- (Conference Schedule)
• https://infocon.org/cons/2600/The Eleventh HOPE (2016)/The Eleventh HOPE (2016) - Hacking Machine Learning Algorithms.mp4
• https://www.youtube.com/watch?v=5e7Rrz5CJHE

Similar Presentations:

• Black Hat USA 2016 / Applied Machine Learning for Data Exfil and Other Fun Topics
• BSidesLV 2017 / The New Cat and Mouse Game: Attacking and Defending Machine Learning Based Software
• DEF CON 32 (2024) / Machine Learning for N00bs Workshop