“Boston. Brussels. Paris. London. Manchester. These are just some of the cities that have experienced terrorist attacks in the last few weeks or years, resulting in tragic loss of life,” said IARPA Deputy Director Stacey Dixon during her GEOINT Foreword keynote Sunday, the morning after a terror spree struck London.
The attacks, Dixon noted, all took place in urban centers inundated with surveillance video, cell phone imagery, and cell phone video.
“What if we had more warning that something was about to happen just based on the activities and behaviors of the perpetrators,” she said. “Would that be helpful? Very much so.”
Dixon, the former deputy director of InnoVision at National Geospatial-Intelligence Agency, joined IARPA, the Intelligence Advanced Research Projects Activity, in 2016. At GEOINT Foreword, she discussed five GEOINT-related research programs, including DIVA (Deep Intermodal Video Analytics), which develops robust automated activity detection for multi-camera streaming environments such as the cities Dixon named.
DIVA will use data from surveillance, mobile phone, and body-worn cameras to detect unusual activity, whether it’s a person brandishing a weapon or someone continuously circling a particular location. The analytics will also be able to perform re-identification among different camera sources.
Given the limitations of humans, unable to review every minute of massive video troves, Dixon said, “We need to teach the computer a human’s ability to understand scenes with speed and scalability that far exceeds that of a single person.”
What if we had more warning that something was about to happen just based on the activities and behaviors of the perpetrators,” she said. “Would that be helpful? Very much so.” —Dr. Stacey Dixon, Deputy Director, IARPA
As with DIVA, the goal of most GEOINT-related IARPA programs is to leverage artificial intelligence and large quantities of training data to automate what is currently being done manually.
Other programs Dixon highlighted include:
The Aladdin Video Program seeks to automatically analyze mountains of open-source video clips. The better trained the algorithm, the better the machine can distinguish between subtleties and annotate them for future reference. The delivered technologies, Dixon said, have already improved video retrieval rates more than 100-fold and are now being used by multiple government agencies.
CORE3D (Creation of Operationally Realistic 3D Environment) provides timely access to geospatially accurate, automatically created 3D models with real physical properties and a level of detail that includes textures, materials, and objects. “Hotspots of the world emerge all the time, and they’re not always places we have access to or have large repositories of data for,” Dixon said. “If we don’t already have a model for it, by the time it becomes a hotspot we generally don’t have access to it. We want to be able to build models for any locations on the Earth.”
Finder aims to build upon existing research systems to develop technology to help an analyst identify locations. Using existing geo-location technologies, publically available data sources, and automation, the program will provide analysts with a candidate list of locations—specifically obscure, less-visited sites, rather than landmarks and highly populated areas with notable features such as mountains or buildings. The program is in the process of being transitioned to government agencies.
The HFGeo (High Frequency Geolocation) Program aims to develop and prototype technology that will provide a significant improvement in the ability to geo-locate and characterize high-frequency (HF) emitters. Recent advances in high dynamic range receivers, antenna techniques, adaptive signal processing, and ionospheric ray path prediction, along with improved measurement and modeling techniques suggest a dramatic improvement in HF reception and geo-location is possible. HFGeo is one of IARPA’s most mature programs, about to enter its final phase.
Unlike the government agencies it partners with, IARPA does not deploy technologies directly to the field. It also does not perform any research in-house. Instead, the organization defines research problems, including some of the most pressing in the IC, and a program manager works with academic and industry researchers to find answers and solve problems.
“Our research not only solves problems in the Intelligence Community but also contributes to the greater field of science and understanding,” Dixon said.
She referenced an NPR story that highlighted Carnegie Mellon University’s Center for Human Rights Science, which applied tools to cell phone data to reveal human rights violations.
“The tools they were using were developed as part of the Aladdin program,” Dixon said. “We’re really proud that our capabilities are not only making a difference in the Intelligence Community but contributing to the goals of organizations completely outside the IC.”
Dixon also talked about IARPA’s prize challenges, in which the organization invites experts from government, academia, and industry to help solve intelligence problems. Experts submit solutions (bypassing proposals) to IARPA, and those with the best solutions receive cash prizes and bragging rights.
Currently, Dixon told the audience, IARPA has a prize challenge open seeking to improve the accuracy of facial recognition.
If we don’t already have a model for it, by the time it becomes a hotspot we generally don’t have access to it. We want to be able to build models for any locations on the Earth.” —Dr. Stacey Dixon, Deputy Director, IARPA
Further, Dixon said, experts can submit research proposals for IARPA’s Seedlings program, which covers a laundry list of areas of interest for nine- to 12-month studies.
“Essentially, if you have a really good idea that is something you’re not quite sure is going to work, that’s an opportunity,” Dixon said. “Our goal with Seedlings is to bring ideas from disbelief to doubt so we can consider whether it’s something we want to use in a program.”
Finally, Dixon said IARPA is always recruiting, from inside and outside the IC. She encouraged GEOINT Foreword attendees to send their ideas to the research organizations within their agencies. “Keep the ideas coming,” she said.