Description
Date depot: 14 février 2020
Titre: ML-driven Co-Design of Network and Recommendation Algorithms
Directeur de thèse:
Thrasyvoulos SPYROPOULOS (Eurecom)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Systèmes et réseaux
Resumé:
In the future, the majority of people will want to access most content (video, audio, etc.) through a wireless device, such as smartphones, tablets, etc. To keep up with the exploding data demand, wireless networks have been traditionally evolving through (i) more bandwidth (e.g. mmWave spectrum for 5G), (ii) more equipment (e.g. network densification through a multitude of small cells), and (iii) better technology (e.g. Massive MIMO, CoMP, etc.). While these techniques are promising, these often come with significant investment costs, and mostly focus on the radio access link, making the network bottleneck simply shift elsewhere (e.g. to the backhaul links or core servers). Coupled with pessimistic predictions that data demand might already surpass the new capacity by the time it’s in place, it seems we might be reaching the limits of what hardware and technology upgrades alone can offer. Video/audio content comprises the vast majority of the above wireless data traffic. As a result, many recent network optimization proposals are content-centric: (i) edge caching of popular content, to improve access latency and reduce transport/server congestion; (ii) traffic steering for load balancing, to move content over underutilized resources (e.g. small cells, other RATs, different MEC/Fog servers); (iii) traffic steering for energy management, to steer content traffic away from underutilized resources (so that
these can be turned off); (iv) broadcast/multicast delivery for Live TV or User Generated Content (e.g., BBC broadcast, Facebook Live), to utilize the same resources for many users (eMBMS)
Doctorant.e: Costantini Marina