NewVectors Projects

Ant CAFÉ

The Ant CAFÉ (composite adaptive fitness evaluation) project is part of ARDA’s Novel Intelligence from Massive Data (NIMD) program, which has a ten year goal of building highly intelligent computer support for homeland security analysts. The Ant CAFÉ is designed to find evidence that corroborates analyst hypotheses. The system will work on an ongoing, anytime basis: analysts will create investigations, and the system will assemble pieces of evidence to match hypotheses as the data becomes available.

Technically, the NewVectors team is breaking new ground by applying research in swarm intelligence, which mimics behaviors of ants and other social insects, to mining unstructured text. The user interface, which is being developed by the Sarnoff Corporation, helps analysts represent hypotheses in the form of concept maps. Concept maps are graphs where the nodes are nouns and the edges are verbs. The Ant CAFÉ backend then processes data in relation to the concept map. This processing includes clustering document paragraphs, identifying hypothesized relations in the text, and assembling matched relations into larger structures that we call scenarios. Currently, we have implemented the first process, clustering.

The advantage of the swarm intelligence approach is that it is highly decentralized and, potentially, massively parallel in a manner that can handle the enormous quantities of intelligence data.