Link-based Anomaly Detection with Sysmon and Graph Neural Networks
Bibliographic record
Abstract
Anomaly detection is a challenge well-suited to machine learning and in the context of information security, the benefits of unsupervised solutions show significant promise. Recent attention to Graph Neural Networks (GNNs) has provided an innovative approach to learn from attributed graphs. Using a GNN encoder-decoder architecture, anomalous edges between nodes can be detected during the reconstruction phase. The aim of this research is to determine whether an unsupervised GNN model can detect anomalous network connections in a static, attributed network. Network logs were collected from four corporate networks and one artificial network using endpoint monitoring tools. A GNN-based anomaly detection system was designed and employed to score and rank anomalous connections between hosts. The model was validated against four realistic experimental scenarios against the four large corporate networks and the smaller artificial network environment. Although quantitative metrics were affected by factors including the scale of the network, qualitative assessments indicated that anomalies from all scenarios were detected. The false positives across each scenario indicate that this model in its current form is useful as an initial triage, though would require further improvement to become a performant detector. This research serves as a promising step for advancing this methodology in detecting anomalous network connections. Future work to improve results includes narrowing the scope of detection to specific threat types and a further focus on feature engineering and selection.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.008 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.003 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".