DIREC Catalyst Project · 2026-2027
We combine machine learning with formal logic to build network intrusion detectors that don't just fit the data — they satisfy provable security properties.
ML-based intrusion detectors trained in the lab latch onto accidental patterns, miss novel attacks, and can be fooled by adversaries.
Differentiable logics — logical rules that plug directly into training, guiding neural networks toward robust, verifiable behavior.
We test on real infrastructure with TDC NET, Denmark's largest telco, bridging the gap between theory and deployment.
We offer paid thesis projects for bachelor and master students. Work on real research problems with industry data and cutting-edge tools. Strong students may continue as research assistants.
Build realistic datasets that expose the gap between lab-trained classifiers and real-world network security.
→Encode security knowledge as formal rules and train classifiers that provably behave correctly.
→Enable organizations to jointly train intrusion detectors without sharing sensitive network data.
→Advance the math that makes it possible to train neural networks with logical constraints.
→Explore the project theme that interests you
Reach out to the supervisor on the Team page
Shape a thesis topic together that fits your interests and the project
We welcome students from computer science, data science, cybersecurity, and related fields. Prior ML or formal methods experience is a plus, but curiosity matters more.