Machine-Learning for QoE of video-conference services

The web is today a general purpose platform where people navigate on websites, watch TV, listen to music, play games, and participate in multiparty conferences. Most of the services over the web have embraced encryption, with HTTPS carrying more than 90% of traffic. This has hampered the ability of in-network devices to classify traffic, and assign a proper QoS/QoE class. Novel techniques are hence needed to re-obtain visibility on application traffic.
This applies, in particular, to multiparty online collaboration applications, from online conference, to multiplayer gaming, from high quality video streaming to potentially connected car services. All these services require strict QoS. Fundamental for any QoS approach, traffic classification is the first problem to solve. Packets or flows need to be classified into a proper class so that QoS management can be applied to optimize the delivery of the data. When considering multiparty online collaboration applications, several streams needs to be managed, e.g., video, audio, control, data for interactive communications, background file sharing, etc. All of them need to be assigned to a proper class of service, giving priority to audio and video streams over background transfers.
Fueled by the recent ability to collect, store and process large amounts of data, big data approaches are changing the way in which problems are tackled, with machine learning solutions that are paving the way to solve problems following a data-driven approach. Machine learning techniques can effectively be used for traffic classification, QoE inference and predictions.

The objective of the PhD is to propose a holistic yet practical solution for the traffic classification and management problems of multiparty online collaboration applications. The following targets are identified:
1. Data collection in testbed
2. Design of a traffic classifier with ML approaches
3. Data collection in the wild to verify and refinement of the classifier
4. Design of possible QoS mechanisms to improve QoE for real time apps
Several video-conference tools will be analyzed. The possibility to use ML in conjunction with Deep Packet Inspection approaches will also be considered.

The candidate should have very good programming skills (in particular: python, C/C++), knowledge and skills on machine-learning techniques and data analysis, knowledge on computer networks and ability to interact in an international research environment.