ATD: Active Learning Activity Detection in Multiplex Networks of Geospatial-Cyber-Temporal Data

Databases of relevance to security and threat detection often have complex multimodal structures with geographic information, timestamps, name labels, human behavior patterns, and more linked together. A modern way of organizing such data is in the form of knowledge graphs. Such large information structures can have hidden information. This project will focus on detecting templates of activity in large and complex data structures organized as a multiplex graph. Examples of real-world data include transportation networks, knowledge graphs of geotagged Twitter data, and synthetic graphs that model human activity. The project aims to investigate a subgraph matching problem that can have a large and combinatorially complex solution space. In order to wade through the myriad of information that might have similar patterns to a known activity template, subject matter experts may be called on to provide additional information. This project will develop algorithms that incorporate a human in the loop and are thus called active learning methods. The research will address both exact and inexact subgraph matching. Metrics for subgraph matching will include graph topology measurements (e.g., graph edit distance), timestamp comparisons, similarities in label attributes (e.g., Levenshtein distance), and geographic distances. The project will quantify both template and world graph active learning strategies. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.