Robust detection of malicious behavior in distributed wireless networks.
Final Report Abstract
Our research in this project has mainly focused on two aspects, namely 1. to develop and evaluate methods of executing computation tasks efficiently in a wireless network in such a manner that the usage of wireless resources and the delay incurred by the computation and the communication necessary to achieve it are minimized, and 2. to secure such computation methods against possible malicious attacks. Both of these aspects are vital for any viable approach to distributed detection of malware and attacks in wireless networks: Item 1 is necessary because any detection algorithm relies on data that is distributed in the network and has to be executed in a timely fashion and in such a manner that it does not impair the main task of the wireless network with excessive overhead resource usage. Item 2 is necessary since for any detection algorithm which is intended as a main line of defense against potential attackers, it is of particular importance to secure the detection algorithm itself against malicious attacks. We have developed several methods for executing communication tasks highly efficiently and scalably in a distributed fashion exploiting the superposition property of the wireless channel. • We have developed distributed versions of two different, widely used iterative methods for computing eigenvalues of matrices. The distributed eigenvalue computation schemes are suitable for large-scale networks and less susceptible to single points of failure than their centralized counterparts. • We have proposed and analyzed methods for computing linear and nonlinear functions over a wireless channel. Applications include the distributed computation of the labeling functions of a class of support vector machines as well as ML schemes based on Boosting, for anomaly detection as well as other tasks. • We have published results on channel resolvability and used them as tools to secure distributed function computation schemes against eavesdropping with the help of a friendly jammer. With these results, we have made significant progress towards the goal of developing defenses against malicious attacks in wireless networks. Due to the general nature of our results, they are promising for many applications within and without the original scope of this project and have sparked many new research directions in our group.
Publications
-
Decentralized eigenvalue algorithms for distributed signal detection in wireless networks. IEEE Transactions on Signal Processing, 63(2):427–440, 2014
Federico Penna and Slawomir Stańczak
-
Energy-efficient classification for anomaly detection: The wireless channel as a helper. In 2016 IEEE International Conference on Communications (ICC). IEEE, 2016
K. Ralinovski, M. Goldenbaum, and S. Stańczak
-
MAC resolvability: First and second order results. In 2017 IEEE Conference on Communications and Network Security (CNS), pages 560–564. IEEE, 2017
Matthias Frey, Igor Bjelaković, and Slawomir Stańczak
-
Resolvability on continuous alphabets. In 2018 IEEE International Symposium on Information Theory (ISIT), pages 2037–2041. IEEE, 2018
Matthias Frey, Igor Bjelakovic, and Slawomir Stanczak
-
Over-the-air computation in correlated channels. IEEE Transactions on Signal Processing, 69:5739–5755, 2021
Matthias Frey, Igor Bjelaković, and Slawomir Stańczak
-
Over-the-air computation in correlated channels. In 2020 IEEE Information Theory Workshop (ITW). IEEE, 2021
Matthias Frey, Igor Bjelaković, and Slawomir Stańczak
-
Towards secure over-the-air computation. In 2021 IEEE International Symposium on Information Theory (ISIT), pages 700–705. IEEE, 2021
Matthias Frey, Igor Bjelaković, and Slawomir Stańczak
