Fraud Detection with Graphs

Detection Methods

In the realm of cybersecurity, graph-based techniques are proving invaluable for detecting malicious activities. Šimon Mandlík explains how these methods can identify harmful domains by analyzing their relationships within vast datasets, reducing the workload for analysts by focusing on the most suspicious domains 1. This approach uses hierarchical multi-instance learning (HML) to represent network activity as graphs, offering scalability where traditional graph neural networks fall short 2.

You can run your model and take the top, say 500 domains daily and send it to your analysts and you can largely reduce the chunk of work they have to do.

--- Šimon Mandlík

By leveraging these techniques, cybersecurity professionals can more effectively pinpoint and address potential threats.

   

Industry Impact

The transition of new cybersecurity methods into industry is marked by significant improvements over existing systems. Šimon Mandlík highlights the collaboration with Cisco, where these techniques were tested on real-world data, showing a threefold increase in efficacy compared to previous algorithms 3. Open-source libraries like JSON Grindr and MILL JL facilitate the adoption of these methods, allowing others to process JSON data and build models efficiently 4.

We have two main libraries which are open sourced completely. They are written in Julia language which is very similar to Python, so anybody can try them.

--- Šimon Mandlík

These advancements not only enhance accuracy but also promote explainability, a crucial aspect in cybersecurity.