Unstable Log Sequence Anomaly Detection - Introducing Vocabulary-Free BERT

Explore an innovative sequence anomaly detection method called VoBERT in this 41-minute Black Hat conference talk. Delve into the challenges faced by Security Operations Centres (SOC) due to overwhelming false positives and the limitations of current analytics models in adapting to unstable log data. Learn how VoBERT improves upon traditional BERT (Bidirectional Encoder Representations from Transformers) models by adding resilience and accurately classifying unstable logs that would typically be considered out-of-vocabulary. Gain insights from speakers Eduardo Barbaro, Daan Hofman, Anna Lukina, and Yury Zhauniarovich as they present this cutting-edge solution for more efficient log sequence anomaly detection in rapidly evolving security environments.

VoBERT: Unstable Log Sequence Anomaly Detection: Introducing Vocabulary-Free BERT