ls -t ~/vasilis/research/selected-work
Selected work
Machine learning and inference on data that never shows the whole picture.
SIGCOMM 2026
supervised learning
Finding the IPs that hide many users
Classifiers trained on global CDN request logs spot carrier-grade NATs, VPNs and proxies, checked against a second model built on public M-Lab data.
read the blog post →
IMC 2024
recommender systems
metAScritic: topology as a recommendation problem
Matrix completion predicts the peering links that BGP and traceroute never see.
NSDI 2019
probabilistic inference
ProbLink: who pays whom on the Internet
Probabilistic inference over noisy BGP paths replaces brittle heuristics for inferring business relationships between networks.
Kepler
anomaly detection
Catching outages from public BGP data
Found 159 outages across 87 facilities and 41 exchange points in four years of routing data.
read on RIPE Labs →
GLOBECOM 2019
random forests
Profiling IoT botnets through DNS
Random Forest classifiers trained on DNS traffic features detect Mirai-like botnet activity in the wild.
2019 to 2020
deep learning
Neural precoding for video
CNNs that prepare video before AVC, HEVC, VP9 and AV1 encoding, cutting bitrate and transcoding cost.