Automation, artificial intelligence and machine learning have all become buzzwords surrounding broad initiatives at the Department of Defense and throughout the intelligence community. But how can these be leveraged in a practical sense to give the military an operational advantage over adversaries?

Jason Matheny, director of the Intelligence Advanced Research Projects Activity, pulled back the curtain during an event hosted by the Defense Writers Group and spoke on several video and imagery projects the IC is funding to get at war-fighter needs:


Finder seeks to geotag images as they are taken, even if the original user disabled the geotag feature. This type of tool is valuable for finding terrorists or even other actors around the globe that may post photos in web forums or social media. The program works by using features in the image — such as skylines, geologic, botanical or architectural — that can pinpoint where on Earth the photo was taken.

Deep Intermodal Video Analysis (DIVA)

DIVA, which is about a year old, automatically detects activities within videos, such as people on security cameras acting suspiciously.

As an example, Matheny said there were video cameras in the Navy Yard shooting that took place in 2013, but there weren’t enough human eyeballs to watch them and sound the alarm. IARPA’s activities in programs such as DIVA seek to automatically detect activities and flag humans.


Aladdin, which has transitioned to the Air Force, can summarize what’s happening in an internet video. Matheny explained that the state of the art a few years ago meant programs looked for the keywords users manually typed in, or “fat fingered,” to tag a video. Aladdin uses AI.

“That doesn’t work, though, if you’re looking at raw video from a UAV or if you’re looking for videos posted to, say, YouTube by a terrorist group,” he said. “Terrorists are not polite enough to tag their own videos.”


CORE3D can generate 3-D models of objects or towns from overhead imagery.

When Navy SEALs raided a compound in Pakistan and ultimately killed Osama bin Laden, it took weeks to create models, Matheny said. CORE3D seeks to build these models in minutes.

Mark Pomerleau is a reporter for C4ISRNET, covering information warfare and cyberspace.

More In Intel/GEOINT