Driving distributed machines learning by the network in IoT

Ph.D proposal in Computer Networks

Context

One research topic developed by the Network and Telecommunication team concerns the Internet of Things (IoT), within each device connected to the Internet has its own computing power and energy capacity. To meet the application QoS requirements, devices can embed different communication technologies, such as the LoRa for long range transmissions and the NB-IoT/LTE-M for low latency communications. In the context of Machine Learning Algorithms (ML), communications has a critical impact on the algorithms performances. Not well considered, this issue is emphasis in wireless networks, due to a variable throughput, packet loss, and power consumption constraints. Depending on the communication technology, additional ones have to be considered such as the duty cycle in LoRa. To ML algorithms requirements in wireless networks, the use of distributed ML can fill this gap.

Research project

The proposed topic of this Ph.D thesis is the driving of distributed ML thanks to the network. Could be considered as a resource allocation problem in a distributed environment, four issues have been identified. The first one is the optimisation of data transmission, through the transmission parameters selection problem with the possibility to slice or aggregate data. The second one concerns the transmission scheduling problem under multiple communication technologies and related to the learning model. The aim is to perform a communication technology selection according to the constraints, such as the energy consumption for a transmission or the desired latency to transmit data for contributing to a global model. The third one is related to the concurrent scheduling, such a medium access technique. The last one is on the convergence and the fairness of asynchronous model. The graph labelling technique will be used to model and guarantee to compute a solution either fair or egalitarian. The mobility of nodes could also considered under a test scenario related to industrial or vehicular networks.

Your profile

We are looking for a new member with a wireless networking and machine learning background and a Master's degree in Electrical or Computer Engineering, Computer Science, or a related field. Ideally, we are looking for someone with an experience, knowledge and interests in: Besides, you should also be able to think critically, have strong English communication skills (oral and written), and enjoy collaboration.

How to apply

We look forward to receive your application by emails (📧) with the following documents : The email must be sent to the three supervisors (benoit.hilt@uha.fr, ismail.bennis@uha.fr and sebastien.bindel@uha.fr) with the following subject: DDML PhD Application from “First-name” “FAMILY-NAME” (this last written in UPPERCASE). As a result, your application email must contain three files.

The Email template is available 👉 📨

⏰ The application deadline is the 6th April 2026.