Physics Monthly -- Topic: Adversarial Attacks

Europe/Berlin
Zoom (Online)

Zoom

Online

Description

Exchange of Topics, Use-Cases and Projects at the Intersection of Physics and Machine Learning

Registration
Participants
    • 10:00 10:05
      Welcome & Opening 5m
      Speaker: Dr Jens Buß (Lamarr Institute, TU Dortmund University)
    • 10:05 10:35
      Enforcing Fundamental Relations via Adversarial Attacks on Input Parameter Correlations 30m

      Adversarial Examples have been shown to be applicable in data augmentation domains on many occasions. However, they are rarely studied in the context of High Energy Physics, as well as in the context of tabular data. To combat this, we have conceptualized a novel adversarial attack algorithm - called the Random Distribution Shuffle Attack - which aims to generate adversaries fooling deep neural networks, while minimizing the perturbations to the individual features' one-dimensional Distributions.

      Speaker: Timo Saala (Lamarr institute, University of Bonn)