Bluff: Interactively Deciphering Adversarial Attacks on Deep Neural Networks

TL;DR

Bluff is an interactive visualization tool that helps researchers understand how adversarial attacks deceive deep neural networks by comparing activation pathways of normal and attacked images.

Contribution

It introduces an open-source, web-based system for visualizing and analyzing the internal mechanisms of adversarial attacks on vision neural networks.

Findings

Reveals attack mechanisms through activation pathway comparison

Enables interactive exploration of attack effects

Facilitates better understanding of model vulnerabilities

Abstract

Deep neural networks (DNNs) are now commonly used in many domains. However, they are vulnerable to adversarial attacks: carefully crafted perturbations on data inputs that can fool a model into making incorrect predictions. Despite significant research on developing DNN attack and defense techniques, people still lack an understanding of how such attacks penetrate a model's internals. We present Bluff, an interactive system for visualizing, characterizing, and deciphering adversarial attacks on vision-based neural networks. Bluff allows people to flexibly visualize and compare the activation pathways for benign and attacked images, revealing mechanisms that adversarial attacks employ to inflict harm on a model. Bluff is open-sourced and runs in modern web browsers.