Air Force Research Laboratory leverages Squadron Innovation Funds to develop Human-Centered Data Analytics Environment

WRIGHT-PATTERSON AIR FORCE BASE, Ohio – Researchers at the Air Force Research Laboratory’s 711th Human Performance Wing here are utilizing a new funding authority to begin designing and building a Human-Centered Data Analytics Environment that will help them store and operationalize huge amounts of both research and operational data.

The volume of data collected during Air Force research endeavors and operations is vast, and human analysis alone is too slow to achieve the decision speed required to win in the information age of warfare.

In February, Chief of Staff of the Air Force Gen. David L. Goldfein announced that Air Force units would receive a new funding authority to accelerate implementation of the National Defense Strategy, with the funds being used for Airmen-led innovations that increase readiness, reduce cost, return time back to Airmen or enhance lethality of the force, according to an Air Force news story.

In Fiscal Year 2018, the Air Force distributed $64 million in Squadron Innovation Funds across the service to reduce resource barriers that prevent ingenuity. Air Force Materiel Command executed more than $4.6 million in Squadron Innovation Funds during the fiscal year, with the 711 HPW receiving $250,000 to launch their Human-Centered Data Analytics Environment project.

“It is well understood among Air Force leadership that data is a growing strategic asset, and we need to take advantage of our own data,” said Dr. Ryan Kramer, Explainable AI lead within the 711 HPW’s Airman Systems Directorate.

Kramer also leads a Commander’s Research Development Funds program that is looking at the integration of analytics, machine learning, and AI approaches for multiple use cases, not just at the 711 HPW but across several AFRL technology directorates.

In data analytics, 80 percent of the work you do is actually getting the data and structuring it in a way that you can run models on, Kramer emphasized.

“We do that on the machine learning and AI side, where we’re trying to formulate complex models that describe complex use cases,” said Kramer. “The first step of running data analytics and machine learning is fusing together all available informative data. It typically exists in stovepipes across the Air Force enterprise, whether it be operational data, research data or open source data that resides in online databases that all potentially inform some of the more complex questions we’re trying to answer.”

According to Todd Overman, who oversees many of the project’s business aspects as civilian executive officer at the 711 HPW, the innovation funds were geared to understand how to take advantage of that.

“We deal with a lot of human data, and there are rules covering how this data is handled and protected,” Overman added. “We have to be cognizant of that as we design our data environments. Our data includes a lot of personally identifiable information and protected health information, so we must ensure proper protections are in place for any new system.”

Researchers from across AFRL are currently working together to analyze software that has already been developed so they don’t have to go back and “reinvent the wheel,” Kramer said, adding, “But really, the most important thing is how do we take those open architectures and ensure that they meet our security requirements?”

There are two components to the efforts for this project according to Kramer, who earned his degree in molecular genetics and bioinformatics.

“First, we are building a data architecture environment that can handle and fuse multiple streams of data on-demand in a data lake model,” he said. “That enables us to combine operational data and research data together to assess its immediate value. Second, we don’t want to just build that architecture. We want to be able to prove it out and show value with very specific use cases.”

In support of the second component, Kramer and his fellow researchers at the 711th are looking at suicide risk prediction.

“We’ve been supporting programs like the Air Force initiative Task Force True North, where the goal is to promote Airmen resiliency by using more proactive approaches in providing mental health support within units. Specifically, the 711 HPW project is studying the risk of suicide within Air Force active duty units,” said Kramer.

“In our initial efforts, we developed a risk stratification model that looks across the active duty population and says, ‘Ok, these squadrons over here at this base are at higher risk for suicide than these squadrons over here at this base,’” Kramer continued. “When we did that, we typically built models every twelve months or so. We would go get our data, pull it down, construct a model, and perform that risk stratification. Then we could feed that information back to Task Force True North or the Integrated Operational Support team where they could intervene with a specific squadron in a precision medicine approach.”

When running these types of analytics, it often takes six or more months to get new data, and researchers don’t always have insight into potentially emerging problems that are happening in cycles of months or weeks. The overarching goal is to reduce these analytic cycles to the point where leadership and interventional teams can have awareness of emergent problems in real time. For suicide risk prediction, working with outdated models can result in missing potentially life threatening circumstances that might be able to be identified with advanced analytics.

As far as where the data analytics environment should be housed, Kramer said researchers believe it should connect to the high-performance computing system at Wright-Patterson.

“These complex machine learning models typically need a lot of compute power. If I have a terabyte of data and want to gain insights to it, I’m going to need that type of compute power to be able to run my methodologies and analytics to get that answer back,” he said.

So far, the 711 HPW has brought one new employee on board to start doing the autonomous modeling to understand what the overall architecture of the analytics environment should look like.

Kramer summed it all up saying, “Data is growing exponentially in every organization and our ability to exploit it is largely growing linearly. Data is a strategic asset, and if you’re not taking advantage of your own data, you’re falling behind.”