Disputasen vil foregå via live streaming i Zoom. Zoom-verten vil moderere det tekniske rundt disputasen, mens disputasleder vil lede selve disputasen.
Ex auditorio spørsmål: Disputasleder vil invitere til at publikum kan stille enten muntlige eller skriftlige ex auditorio spørsmål. Du kan etterspørre dette ved å trykke på "participants" etterfulgt ved å klikke "Raise hand".
Prøveforelesning
Generative Adversarial Network (GAN) and their application to Computer Network Security
Hovedfunn (på engelsk)
Over the last decade, the number of mobile devices has increased and Wi-Fi networks have become the main access medium for delivering broadband internet in homes and buildings. Meanwhile, internet service providers (ISPs) have promised to deliver carrier-grade broadband internet by deploying fiber backhaul and customer premises equipment (CPE) operating at the latest 802.11 standards. Hence, end-users expect uninterrupted and high-quality internet access anywhere in homes and buildings particularly during pandemics when home schools, home offices and on-demand entertainment services have become the norm. However, the best-effort transmission mechanism implemented in wireless equipment may hinder users from experiencing carrier-grade internet services in homes and buildings. The fact that the perceived quality of internet services significantly affects user satisfaction underscores the need for monitoring quality of service (QoS) parameters to assess the perceived quality of service (PQoS) within customer premises. This thesis studied existing monitoring and management solutions and analyzed the quality requirements of existing and novel services. On this basis, this thesis proposes a novel PQoS assessment using machine learning (PQoSML) methodology to classify the perceived quality of internet services and Wi-Fi networks using machine learning (ML) techniques. In the proposed methodology, ML techniques produce an interpretable ML model that accurately correlates network performance parameters to the perceived quality of individual internet services or the entire Wi-Fi network. The computationally efficient ML model, which is executable in the form of edge computing on off-the-shelf Wi-Fi access points (APs) on customer premises, reduces the cost of data analytics while preserving customers’ privacy. In effect, the PQoSML methodology enables ISPs to replace traditional QoS monitoring with an efficient wireless quality assessment (QA) strategy and thus to deliver predictable and measurable wireless quality.
For mer informasjon
PhD rådgiver, Ida Elisabeth Rydning, i.e.rydning@its.uio.no
Teknisk ansvarlig, Arild Hemstad, arild.hemstad@its.uio.no