10:29 am May 8th, 2008 by wkallander

Today, we talked about using anomaly detection techniques for the problem of hidden community discovery. The idea is to use “event explanation” as a filter out normal (explained) events and then look at the largest signal that remains. By explaining the normal activity away, what remains is the anomalous activity which is the subject of analysis. One issue that stands out is the typical network analysis techniques make assumptions about communications channels (i.e. ignoring out-of-band relationships).

This reminds me of the difference between “open world” versus “closed world” models. Briefly, most relational databases make the closed world assumption - that if no data is present to support a particular relationship, then it is concluded to be false (or non existent). Conversely open world models (like that adopted by OWL inference), assumes incomplete knowledge (or unobserved relationships), so that if a relation does not exist in the graph, then it is not be inferred to be non-existent, just absent or unobserved.

Lastly, Professor Wu reiterated the point that I made as a comment on my Krebs review, that we need to formally or more rigorously define the model of the problem of discovering terrorist networks.

This entry was posted on Thursday, May 8th, 2008 at 10:29 am and is filed under ECS289M. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.