Detecting anomaly patterns in networks

Nowadays, we are surrounded by a large volume of data. They are an integral part of our everyday activates such as social network interactions, banking data, health monitoring, marketing and so forth. A large portion of this data is generated from a network; for instance Facebook data, fMRI measurements in brain networks, and transportation data. The complexity of the network on which these data reside renders direct signal processing a challenging task. Graph signal processing is a novel research area that aims at exploiting this network structure to efficiently analyze the data at hand.

A key task in network data is to detect anomalies with respect to a normal network behavior. These anomalies can be fake news in a social network, Alzheimer diagnosis in brain network, or traffic bottlenecks in transportation networks. An efficient solution to this task is highly appreciated in both the academic and industry works since they are directly linked with practical challenges.

Assignment

The problem formulation for anomaly detection can be easily extended from classical detection theory, whereas the challenge is in including the network structure into the play. There are some current works that have dealt with this task and similar results can be adopted as well. The Neyman-Pearson and Bayesian detection theories open the doors to achieve optimal detecting performance where the signal of interest is cast as a vector variable.

In this project, your task would be to derive an accurate model for detecting anomaly patterns in networks. You further need to derive a detector for the considered problem and analyze its performance. You will work with both simulated and real data and you are requested to compare the performance with other state-of-the-art approaches.

Requirements

The project requires a self-motivated student who has a strong linear algebra and statistical signal processing background, and is interested in theoretical and practical aspects of anomaly detection on networks. Furthermore, you are required to respect deadlines in delivering the results. You should have advanced coding skill in one computer language, including Matlab, Python, or R.

Contact

prof.dr.ir. Geert Leus

Signal Processing Systems Group

Department of Microelectronics

Last modified: 2018-03-21