Job Description

PhD director: E. Veronica Belmega (UGE, ESIEE Paris)
Co-advisors: Romain Negrel (UGE, ESIEE Paris), Anne Savard (IMT Nord Europe)
Collaborators: Giacomo Bacci (University of Pisa, Italy), Panayotis Mertikopoulos (CNRS)

1. PhD scientific context and objectives

AI and machine learning have rapidly become essential enablers for 6G future wireless networks. In this context and motivated by the highly dynamic nature of emerging wireless networks (B5G, IoT), online optimization methods have been successfully exploited to design resource allocation policies in various problems by allowing for temporal variations that are completely arbitrary (even non stationary), thus being particularly suitable in accounting for non-random fluctuations in the wireless medium, unpredictable connectivity and mobility behavior of IoT devices.

Most existing works focus on designing first order online algorithms solving convex online optimization problems. Such algorithms require the receiver to feedback the past gradient information to the transmitter, which can be very costly (massive MIMO) and also problematic in low-complex low-energy IoT networks. Also, maximizing the energy efficiency of IoT networks often leads to non convex and difficult optimization problems, when incorporating practical aspects. Solving such complex problems will require modern machine learning tools such as deep neural networks to be cleverly combined with online and reinforcement learning to cope with the dynamic wireless environment and mobility aspects.

The main objectives of this PhD project are:
OBJ1. Design efficient online optimization algorithms requiring low-cost and energy-efficient communication feedback (one scalar or even one bit);
OBJ2. Design online algorithms maximizing non convex energy efficiency optimization problems exploiting and combining modern deep learning techniques and non convex online optimization.

Zeroth order online algorithms based only on a scalar value of the objective coupled with a one-point stochastic gradient estimation lead to very poor algorithm efficiencies in reaching the optimal performance. By using a callback trick to build a two-point gradient estimation, we can recover the same efficiency as the first order algorithms in the static optimization case. The challenge in OBJ1 is to exploit the callback trick in the online setting and recover the optimal efficiency.

To tackle the non convexity of energy efficiency optimization problems in unpredictable IoT networks, the challenge in OBJ2 is to exploit non convex online optimization techniques coupled with advanced AI techniques based on deep learning combining their strengths.

The obtained results will be published in top-tier venues (research journals and international conferences) such as: IEEE Trans. on Signal Processing, IEEE Trans. Machine Learning in Communications and Networking, IEEE Trans. On Wireless Communications, IEEE ICC, IEEE GLOBECOM etc.

2. PhD environment

The 3-year PhD candidate is funded within the PEPR 5G project on the "Development of advanced technologies for 5G and future networks”.
The PhD candidate will receive a gross salary of around 2k euro per month.
The PhD candidate will enroll in the Université Gustave Eiffel (UGE) doctoral school and will have to pay a relatively modest registration fee (around 400-500 euro per year).
The PEPR 5G project budget includes further PhD related expenses such as: participation to international conferences, short mobilities, publication fees, portable PC, etc.

3. How to apply?

The applicants should hold a Master degree (BAC+5) and have a strong background in either electrical engineering or computer science or applied mathematics. A good English level in writing, reading and speaking is required. Finally, having a strong mathematical background and/or computer literacy skills (Python, C++, MatLab, etc.) is a definite plus.

Applications will be received via the online form below until the position is suitably filled.
NB: no applications will be received via email.

The application dossier should include: a short motivation letter (1 page max), an academic oriented CV (2 pages max), the academic track records for the M.Sc. and B.Sc. (post-BAC)
including rankings, and two relevant reference letters.

Online application form: https://forms.gle/MGj9xf6xBYE5GJpa6
Contact: romain.negrel@esiee.fr, anne.savard@imt-nord-europe.fr

Follow @eurasip