Cyber Security Webinar by Daniël Vos Msc - Training Robust Decision Trees Against Adversarial Examples

Join Zoom Meeting

https://tudelft.zoom.us/j/94484315617

Meeting ID: 944 8431 5617

Passcode: 049126

Abstract 

Decision trees and tree ensembles are among the most popular machine learning models for tabular data but they suffer from adversarial examples. When we make specific changes to a model's input, the resulting prediction can be completely wrong. Since this finding, researchers have proposed different methods to train trees that are robust so such attacks. However, these methods can take hours to run and offer no performance guarantees.

In this seminar we will look at how to speed up training by orders of magnitude and explain how we can prove the optimality of robust decision trees.