Poison in the Wires: Interactive Network Visualization of Data Poisoning Attacks

No ratings

Presented at No Hat 2025 by

Maria Khodak

What if we could not only visualize poisoned training data, but interact with it? As machine learning models become increasingly integral to software applications, they also face increased risks from adversarial attacks, notably data poisoning. After experimenting with data poisoning, I noticed that training data fed into machine learning models resembled graphical information that could be defined using a visually represented network. This observation inspired me to create GraphLeak, an open-source, interactive web tool designed to visualize how poisoned training data alters network structure. We will explore how adversarial data manipulation impacts graph-based representations. Building on network science concepts, this session will go deeper: not just showing how poisoning affects structure, but allowing users to directly interact with both poisoned and untouched datasets in real time. I explore data poisoning with several notable examples, as well as tested case studies. In addition to exploring network visualization through GraphLeak, I will demonstrate a live data poisoning attack, collect the poisoned data, and then visualize this attack with GraphLeak. My session aims to offer hackers and cybersecurity professionals a visual and graphical perspective on data poisoning vulnerabilities. By visualizing data poisoning, we can more easily detect tampered data.