Army-Funded Academe Team Creates, Tests New AI Learning Framework

Jeff Brody

Researchers from North Carolina State University have developed and tested a framework to support the learning capacity of artificial intelligence platforms under a project with the U.S. Army. The Army-funded Learn to Grow effort aims to reduce the cases of AI forgetting about learnings, the service branch said Monday. The effort's team also demonstrated the use of the framework to help an AI system better perform previous tasks.

"We expect the Army's intelligent systems to continually acquire new skills as they conduct missions on battlefields around the world without forgetting skills that have already been trained," said Mary Anne Fields, program manager for intelligent systems at Army Research Office.

According to the framework, raw data enter deep neural networks and exit as resulting task outputs. These networks contain multiple layers that each tailors input data for specific tasks.

"We've run experiments using several datasets, and what we've found is that the more similar a new task is to previous tasks, the more overlap there is in terms of the existing layers that are kept to perform the new task," said Xilai Li, a co-lead author of the project's corresponding paper.

The team also compared Learn to Grow's results with other AI learning methods, with findings that indicate the new framework's edge in accuracy over the others. The National Science Foundation supports the Learn to Grow effort.

You may also be interested in...


DOD Opens Cloud Database on Radiation-Hardened Microelectronics; Rich Ryan Quoted

Rich Ryan, director for international programs, nuclear forensics, resiliency and survivability in the office of the deputy assistant secretary of defense for nuclear matters, said the U.S. military has opened a cloud-based library to support the protection of microelectronics used in nuclear systems. The Department of Defense handles strategic deterrence through its nuclear triad and space systems, whose electronic components can be vulnerable to ionized radiation, electromagnetic pulse and other threatening types of stimulus.