ADELPHI, MD. -- As the Army adapts to the changing technological landscape of the modern world, the need for greater insight into network science has only grown significantly over the years.
Whether in setting up autonomous systems to aid the warfighter or synergizing operations across multiple domains, Army researchers have persistently tried to understand the information throughput of communication networks that govern the battlespace.
Now, one of those longstanding efforts is reaching its conclusion after a decade of promising research and partnerships between Army laboratories, private industry and academia.
For almost 10 years, the Army's Network Science Collaborative Technology Alliance has served as a vital junction for the field of network science, bringing together leaders in computer science, information science, social science, social networking and telecommunications into one multidisciplinary research group.
The research program was spearheaded by the U.S. Army Combat Capabilities Development Command's Army Research Laboratory and included industry collaborators such as IBM and Raytheon BBN Technologies as well as universities such as the University of Illinois at Urbana-Champaign, Rensselaer Polytechnic Institute, Penn State and many others.
As one of the laboratory's collaborative technology and research alliances, the NSCTA established partnerships between the Army and various industrial and academic organizations to address technical challenges posed by interacting networks within the Army's Multi-Domain Operations and develop approaches to predict and control their behaviors.
Among the numerous interdisciplinary technical areas, or thrusts, that guided the alliance's research, one of the targets was Information Processing Across Networks for Decision-Making, which investigated different strategies behind information discovery and analysis.
Dr. Lance Kaplan, the Government Technical Area Lead for the IPAN thrust, described some of the goals.
"We focus on leveraging indicators within the social and cognitive spheres and within information and communication networks to obtain situational awareness for the warfighter," Kaplan said. "Part of our work involves improving machine learning by exploiting network context and addressing some of the unique challenges that the Army has that you wouldn't necessarily have places like Google exploring."
For example, the IPAN team is keen on tackling situations where warfighters have to employ complex networks despite limitations in data quantity, time, and bandwidth. While companies like Google can feed its networks endless amounts of data to train its machines, warfighters may find themselves in a disadvantaged network that may hinder how they make crucial decisions.
"Soldiers at the network edge may have limited access to data or bandwidth to reach the Cloud, so they need to think about what to download on their own systems ahead of time and how much," Kaplan said. "They don't have the luxury of time to train up their translation systems, their search engines and things like that. What we need to understand is, how we can get systems to still perform well even with limited data trained on them."
In light of this dilemma, researchers within the thrust focused heavily on developing techniques that will let them train systems with a lot less data.
One of the key research components of interest in this regard was knowledge network construction, which aimed to transform any and all structured or unstructured data from different domains into graphs of entities and events connected by various relationships. Through this approach, the researchers could utilize implicit, or hidden, information held within the structural representations within languages.
Over the ten-year span of the alliance, the IPAN team examined different methods of discerning information from context that combine high-level human- readable ontologies and annotated data (symbolic representations) with quantification of semantic similarities of linguistic items (statistical representations) to map structured information concepts into a knowledge graph that represents the extracted concepts.
Through a combination of distributional semantics, graph-based analysis and zero-shot learning, the IPAN team explored countless different ways to push the boundaries of machine learning for information extraction.
"While we are using techniques developed by others for certain applications, there are many other ways where we are leading in terms of pushing the state of the art," Kaplan said. "As a result of the NSCTA, we are one of the leaders in graph-based machine learning."
In addition to knowledge network construction, researchers within IPAN also directed their attention toward what they have called querying knowledge networks, which concentrates on training information processing systems to more efficiently deliver specific information upon request.
According to Dr. Clare Voss, one of the lab's computer scientists working within the IPAN thrust, how the user interacts with the knowledge networks can greatly influence mission success.
"Oftentimes, when you pose a question to a knowledge network, you expect to receive an answer back," Voss said. "From the perspective of a military commander, you don't always know if the answer is in the knowledge network, but you do want the analyst who is trying to answer the commander's questions to be learning background knowledge from the network as they go. So rather than relying on a single question-answer approach, we're taking into account when the question that the user starts with needs to be refined. This may mean that their question is reduced to a set of simpler questions or enhanced with parts of the question augmented by the system's inference component."
By considering the needs of the user, the researchers within IPAN can structure the network to ease the mental burden of the warfighter when it comes to decision-making.
With the NSCTA reaching its conclusion, Kaplan views the research program as an essential launch platform for future advancements in artificial intelligence and machine learning. However, he believes that there is still much work remaining in order for the Army's networks to function ideally.
"Information networks like Google do a pretty good job of finding the information you need, but there are times when you are looking for something very specific and it takes a long time to figure out what the key words are to go ahead and find it," Kaplan said. "Now imagine needing to find that information right away. For the military, improvements in network systems are things that can have life or death-type implications. A lot of important decisions are made based on the information available and it is crucial to supply the decision maker with the best and most relevant information."
While the NSCTA will soon come to a close, the partnerships that flourished over the decade will continue to endure in order to aid the warfighter with innovative technologies.
As a result of this research program, the laboratory has not only enhanced the Army's understanding of network science, but it has also fostered vital connections with scientists in industry and academia for future collaborations.
The CCDC Army Research Laboratory is an element of the U.S. Army Combat Capabilities Development Command. As the Army's corporate research laboratory, ARL discovers, innovates and transitions science and technology to ensure dominant strategic land power. Through collaboration across the command's core technical competencies, CCDC leads in the discovery, development and delivery of the technology-based capabilities required to make Soldiers more lethal to win our Nation's wars and come home safely. CCDC is a major subordinate command of the U.S. Army Futures Command.
U.S. Army CCDC Army Research Laboratory
U.S. Army Combat Capabilities Development Command