In the future, billions of IoT devices and sensors will be connected and cooperate together to detect events in distributed monitoring and alert systems in applications like intrusion detection, target tracking, smart grids and smart homes, industry 4.0, e-Health or intelligent transportation (connected cars or drones). Many of these applications have system-critical security requirements, in the sense that eavesdroppers and intruders should not learn or influence decisions or data and measurements. The objective of this thesis is to characterize the fundamental limits and devise new strategies of distributed detection systems with security constraints against external and internal eavesdroppers that should not be able to learn the data and measurements performed at the sensors. In this goal we shall propose an information-theoretic analysis as well as learning-assisted detection strategies.

In our previous works, we have derived fundamental limits of multi-hop and multi-sensor detection systems without secrecy constraints and of single-hop systems with a security constraintagainst an external eavesdropper. All these works focused on binary hypotheses and asymmetric scenarios where the error probability under the null hypothesis (false-alarm probability) needs to be kept below a certain threshold while the error probability under the alternative hypothesis (miss-detection probability) should decay exponentially fast in the blocklength with largest possible exponent. In addition, in the secrecy setup with an external eavesdropper, the uncertainty (equivocation) about the sensor measurements should stay above pre-defined security thresholds given the two hypothesis.

The goal of this thesis is to extend the single-hop scenario with external eavesdropper in [5] to multiple hypotheses with either symmetric or asymmetric requirements on the probabilities of error under the various hypotheses. The focus is also on multi-hop scenarios where the security constraints are not only with respect to external eavesdroppers but also against system-internal intermediate relaying terminals. In such systems, the exchange of private sensitive information is undesirable with other terminals or decision centers, or even infeasible due to communication and resource constraints. Therefore, it is particularly important to analyze smart device selection strategies to identify which relays should be included in the distributed decision task, in particular in view of the internal security requirements, the overall energy-consumption and the various communication requirements of the system.
To derive the fundamental limits of the described systems, we shall rely on a recent converse strategy that we pioneered by using a change-of-measure argument and by proving asymptotic Markov chains, combined with the tools that we already used in our previous works. Leveraging on these limits and with the aid of deep learning and reinforcement learning techniques, smart device selection strategies and practical secure distributed detection systems will be proposed.

The project consists of three parts : i) characterize the fundamental limits of distributed detection systems in general multi-hop scenarios considering multiple hypotheses with different constraints on the error probabilities and the equivocations measured at external eavesdroppers, ii) extend these fundamental limits to scenarios where the conveyed information (data and/or decisions) need to be kept secret from other system terminals, and iii) design efficient distributed learning strategies to perform selection of devices participating collaboratively in the distributed decision. Distributed secure learning and distributed decision techniques will be designed taking into account the communication rate and security requirements, and the energy and device constraints of future networks.

Thesis Prerequisites:
- Students with a master’s degree or with an engineering degree (or completing their degree)
- Excellent analytic skills
- Solid background in mathematics and probability
- Good knowledge on digital communications, information theory, statistical information processing, and/or optimization techniques for wireless networks
- Programming skills : C/C++, Python, Matlab, Linux.

Thesis Information:
The thesis is jointly supervised by Dr. Mireille Sarkiss, Telecom SudParis and Prof. Michèle Wigger, Telecom Paris. The PhD student will be working on the Palaiseau campus of Telecom SudParis and Telecom Paris and will be enrolled at Telecom SudParis, Doctoral school of Institut Polytechnique de Paris (IP Paris).The thesis project is funded by France 2030 PEPR Future Networks.
- Period : 2024-2027, starting as soon as possible and by October 2024.
- Institutions : Telecom SudParis and Telecom Paris, Institut Polytechnique de Paris, France
- Advisors and contacts:
– Dr. Mireille SARKISS, Associate Professor at Telecom SudParis, mireille.sarkiss@telecom-sudparis.eu
– Pr. Michèle WIGGER, Professor at Telecom Paris, michele.wigger@telecom-paris.fr

References:
[1] M. Hamad, M. Wigger, and M. Sarkiss, “Cooperative multi-sensor detection under variable-length coding,” in 2020 IEEE Information Theory Workshop (ITW), pp. 1–5, 2020.
[2] M. Hamad, M. Wigger, and M. Sarkiss, “Optimal exponents in cascaded hypothesis testing under expected rate constraints,” in 2021 IEEE Information Theory Workshop (ITW), pp. 1–6, 2021.
[3] M. Hamad, M. Sarkiss, and M. Wigger, “Benefits of rate-sharing for distributed hypothesis testing,” in 2022 IEEE International Symposium on Information Theory (ISIT), pp. 2714–2719, 2022.
[4] M. Hamad, M. Wigger, and M. Sarkiss, “Multi-hop network with multiple decision centers under expected-rate constraints,” IEEE Transactions on Information Theory, 2023. https://arxiv.org/abs/2208.14243.
[5] S. Faour, M. Hamad, M. Sarkiss, and M. Wigger, “Testing against independence with an eavesdropper,” in 2023 IEEE Information Theory Workshop (ITW), 2023. https://arxiv.org/abs/2211.03475.
[6] S. Sreekumar, A. Cohen, and D. Gündüz, “Privacy-aware distributed hypothesis testing,” Entropy, vol. 22, 2020.
[7] M. Mhanna and P. Piantanida, “On secure distributed hypothesis testing,” in IEEE International Symposium on Information Theory (ISIT), pp. 1605–1609, 2015.
[8] M. Hamad, M. Wigger, and M. Sarkiss, “Strong converses using change of measure and asymptotic markov chains,” in 2022 IEEE Information Theory Workshop (ITW), pp. 535–540, 2022.