FGSM Explainer 2022

The FGSM Explainer is an on-going project in which we design and implement an interactive web-based visualization software that demonstrates the underlying logic and consequences of FGSM Attack, a gradient-based white-box attack in Adversarial Machine Learning (AML), while simultaneously performs data analytics and model performance analysis.

The primary objective of our study is to explore how visualizations can help machine learning practitioners understand the underlying logic of adversarial attacks. Especially, we aim to explore for practitioners with limited background in AML, how effectively can visualizations highlight and explain the adversarial images, as they tend to be imperceptible to human eyes but can fool state-of-the-art classifiers.

    Core Features
  • An Interactive Visualization of FGSM Attack
  • Provide data analytics and model performance analysis in parallel
  • Demonstrate the underlying logic and consequences of adversarial attacks via animated sequences
© 2024 Yuzhe Y. All Rights Reserved.