Explore online
Use the limited browser trial.
Inspect the research interface with its prepared data and interactive views.
Open trial ↗
U.S. patent application 63/168,686 · Pending
Find what looks different.
VAAD brings UN-AVOIDS together with linked interactive plots so an analyst can examine unusual observations instead of relying on a single hard decision.
Research software and limited trial. It is not presented as a production monitoring service.
Conceptual view of relative anomalousness.
The system
One method, several ways to use it.
VAAD is the visual environment. UN-AVOIDS is the underlying approach for outlier visualization and anomaly scoring.
Use the package
Install UN-AVOIDS from PyPI.
Integrate the available Python package into a research workflow.
View PyPI package ↗Study the code
Read or reimplement the method.
The research code is available on GitHub and Code Ocean. Commercial use requires a quotation.
View source code ↗Science behind VAAD
UN-AVOIDS makes anomalousness visible.
The method is described in the peer-reviewed IEEE research record. These are its central properties.
Unsupervised
It works without prior training or labelled examples, including single-class labels.
Nonparametric
It does not require an assumed probability distribution, and it has no tuning parameter controlling the result.
Built for exploration
The analyst can inspect outliers visually and link the UN-AVOIDS view with matrix plots, parallel coordinates, and other interactive views.
Invariant score
Each observation receives a normalized score between 0 and 1, expressing its level of anomalousness instead of forcing a hard threshold.
Applications
Anomaly detection is not limited to network security.
The same idea applies wherever unusual observations need to be compared with the rest of a population.
Network security
Investigate unusual activity.
Network attacks may produce feature patterns that differ from ordinary network activity.
Disease detection
Examine atypical measurements.
Medical observations can be explored for patterns that differ from the rest of a dataset.
Fraud detection
Flag transactions for review.
Transactions that differ from a wider population can be raised for closer examination.
Research record
Patent and selected publications.
The application remains pending. The publication and reproducibility capsule below form the research record for UN-AVOIDS.
U.S. patent application Pending
“Unsupervised and nonparametric approach for visualizing outliers and invariant detection scoring.” U.S. Patent Application 63/168,686. Status: pending.
Selected publication 1 paper
- Yousef, W. A., Traoré, I., & Briguglio, W. (2021). “UN-AVOIDS: Unsupervised and Nonparametric Approach for Visualizing Outliers and Invariant Detection Scoring.” IEEE Transactions on Information Forensics and Security, 16, 5195–5210.
Code Ocean capsule Reproducibility
The capsule preserves the research code and environment associated with the published method.
Open the capsule ↗Contact
Want to use UN-AVOIDS commercially?
Tell us how you plan to use the method. The email address appears only after you click.