Presented by: Prahar Shah

Date: Thursday, July 2, 2026
Time: 11:15 am
Place: Online via Zoom
Sponsor: Dr. Issa Traore

https://uvic.zoom.us/j/89697968567?pwd=afVPUP3IGegZVLlEHaxk6efg92Hh4C.1

Abstract:

This seminar presents the design and evaluation of a graph-based anomaly detection approach for detecting lateral movement attacks. The project models activity across Linux and Windows hosts as a heterogeneous graph and uses a graph neural network to identify suspicious relationships between users, hosts, processes, files, services, and network connections. The evaluation compares this approach with rule-based detection, Isolation Forest, and LSTM autoencoders under concentrated and low-and-slow attack conditions, showing that graph-based modelling can improve detection coverage, reduce alert volume, and provide useful investigation context.