Data-Driven Learning for Physical Layer Security and Curiosity-based Analysis of Anomalous Traffic for LTE Devices

Interdisciplinary Areas: Internet of Things and Cyber Physical Systems, Data/Information/Computation, Security and Privacy

Project Description

The LTE protocol for cellular networks offers great flexibility and scalability compared to 3G and 2G predecessors. However, the added flexibility comes with a price of extra vulnerability to unauthorized usage patterns. With the incorporation of diameter signaling, new threats exist at both the access and non-access strata. Threat identification and recovery mechanisms are hence needed at all layers of the communication stack. Our focus here is on the physical and network layers. For the former, current approaches are limited by the accuracy of the models used for the channel and device parameters. For the latter, current approaches rely on knowledge of attack types and remain prone to novel threats. The proposed task is two-fold: First, state of the art deep learning architectures will be tuned to identify the imprint of received wireless signals, and theoretical guarantees will be given on the capacity of physical layer secure communication based on the output of the machine learning classifiers. Second, a data-driven anomaly detection approach will be developed for higher layers in the communication stack, followed by analysis of previously unknown attacks using curiosity-based deep reinforcement learning. 

Start Date

July 1, 2019

Postdoc Qualifications

- Ph.D. in Electrical, Computer Engineering or Computer Science
- Strong background in Statistics and state of the art Machine Learning algorithms
- Familiarity with concepts of wireless communications and networking security
- Motivation to engage and lead research in machine learning for wireless network security that include both theoretical and applied aspects. 

Co-advisors

Aly El Gamal, elgamala@purdue.edu, ECE Department, https://web.ics.purdue.edu/~elgamala/

Walid Aref, aref@purdue.edu, Dept. of Computer Science, https://www.cs.purdue.edu/people/faculty/aref/

Arif Ghafoor, ghafoor@ecn.purdue.edu, ECE Department, https://engineering.purdue.edu/ECE/People/ptPeopleListing?group_id=2571&resource_id=2827 

References

1. X. Liu, D. Yang, A. El Gamal, "Deep Neural Network Architectures for Modulation Classification", available at https://arxiv.org/abs/1712.00443

2. T. Shawly, A. Elghariani, J. Kobes, A. Ghafoor, "Architectures for Detecting Real-time Multiple Multi-stage Network Attacks Using Hidden Markov Model", available at https://arxiv.org/abs/1807.09764

3. S. Mavoungo, G. Kaddoum, M. Taha, G. Matar, "Survey on Threats and Attacks on
Mobile Networks", available at https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7547270

4. R. Chalapathy, A. K. Menon, S. Chawla, "Anomaly Detection using One-Class Neural Networks", available at https://arxiv.org/abs/1802.06360

5. Y. Burda, H. Edwards, D. Pathak, A. Storkey, T. Darrell, A. Effros, "Large-Scale Study of
Curiosity-Driven Learning", available at https://pathak22.github.io/large-scale-curiosity/