Objective:

• Given graph-based representation of intelligence data, learn structural patterns distinguishing threat groups/events from non-threat groups/events

Technologies:

• Graph-based data mining using an information compression measure

• Identify interesting subgraphs in an input graph

• Identify distinguishing subgraphs occurring in one graph (threat) but not another (non-threat)

Uniqueness:

• Learning graph-based structural patterns, rather than correlations of features

• Efficient and effective search in the exponential space of subgraphs

PIs: Larry Holder and Diane Cook

Sponsor: DHS/DoD; $500K past 3 years