Research
Cyber & ML
research
Papers and code from our team’s work on security-log analysis, synthetic data and document understanding.
Cyber security
2024
Journal of Network and Systems Management
LLM4Sec
Language models
for security logs
A research pipeline for evaluating language models on application and system logs.
Egil Karlsen and collaborators benchmarked 60 fine-tuned language models across six application and system log datasets. The study compares architectures and examines how fine-tuning affects adaptation to different log types.
The LLM4Sec repository contains the experimentation pipeline used for this research. The code and papers describe the methods, datasets and classification results.
Machine learning
WACV Workshops
2026
SynthForm
Synthetic data for
document understanding
A dataset and models for extracting information from degraded forms.
SynthForm studies form understanding when documents contain scan defects, fax artifacts, blur and compression. These conditions can make text and layout harder to interpret.
Co-authored by Egil Karlsen and presented at a WACV 2026 workshop, the paper introduces the SynthForm-3k dataset and SynthForm-VL models. The paper and repository describe how the data is generated and how the models are evaluated.
Collaboration
Research with Vaktros
We welcome research collaboration in security data, sensor fusion and machine learning.
Contact our team