Contrastive Learning and Correlation Clustering for Sequences of Network Telescope Data
Researchers have developed a new method using contrastive learning and correlation clustering to analyze sequences of network telescope data. This approach aims to identify relationships between internet scanning activities without requiring semantic annotations. A transformer model embeds network flow records, and the learned similarities are then used to solve a correlation clustering problem, yielding clusters that align with scanner labels. AI
IMPACT Introduces a novel approach for analyzing network traffic patterns, potentially improving cybersecurity threat detection.