Abstract

Network traffic classification is essential for ensuring Quality of Service, enforcing policies, or identifying malware, botnets, and their corresponding command-and-control servers. However, the growing complexity of encrypted traffic, due to the prevalence of TLS , QUIC, and DNS over HTTPS , presents significant challenges to traditional traffic analysis techniques. This talk will discuss advancements in machine learning for processing encrypted network data at the flow level in high-speed computer networks, outlining methods that leverage statistical information from metadata such as packet lengths and inter-arrival times. The talk will cover different statistical data representations on input, modelling architectures, and open problems in flow embedding extraction, contrastive learning, and model transfer.

Bio: Daniel Poliakov is a Ph.D. student at Brno University of Technology with current research interests in data modelling, neural networks, and representation learning for applications in network traffic monitoring and security. Prior to starting his Ph.D., he focused on automated malware analysis, network security research, and threat intelligence.

Join online: https://teams.microsoft.com/l/meetup-join/19:meeting_ZDhlNGE0YTAtMDY5Mi00NGZhLTljOGEtM2ZiNGEyY2Y4Nzc2@thread.v2/0?context=%7B%22Tid%22:%2249a50445-bdfa-4b79-ade3-547b4f3986e9%22,%22Oid%22:%22c74ff4ca-98fe-4b28-9889-e119acc12f30%22%7D