What does streamlining Boeing’s airplane manufacturing process have to do with satellite imaging? Both are engineering problems that require sorting through large amounts of data and applying specialized rules to interpret it. And both problems can be solved using the Adaptive Resonance Theory (ART) neural model designed by Boston University professors Gail Carpenter and Stephen Grossberg.

The Department of Cognitive and Neural Systems technology lab develops neural models similar to TELOS but focuses on memory rather than motor control. These models can help solve challenging engineering problems by mimicking the way the brain makes sense of data and images.

At Boeing, engineers design airplanes by creating small parts and systems of parts. Engineers often end up redesigning parts that already exist because sorting through the database of existing designs takes longer than re-creating the part. But the ART model can quickly learn how to retrieve designs that match a rough sketch of a new design, saving time and money that engineers would otherwise have wasted on duplicated efforts. Boeing claims ART has saved them millions of dollars each year.

ART also has applications for remote sensing. Satellites collect images of the Earth as they orbit but lack the brains to make sense of the data, “whereas a human would notice interesting changes and patterns,” says Carpenter. In a collaborative effort with the BU Center for Remote Sensing, Carpenter and geographer Sucharita Gopal have used ART to transform hordes of satellite images into maps that show meaningful interpretations of the way the Earth’s landscape is changing. In fact, after the first Gulf War, the ART system recognized a new pattern that didn't fit into its existing land cover categories: the burning oil fields in Iraq.

Today, Carpenter is extending ART into a distributed model that learns to associate data with multiple categories. For example, in satellite images ART might categorize a pixel, representing a 30x30-meter section of the Earth’s surface, as a conifer forest, but in reality that area might be home to both conifer and hardwood. Since the distributed ART model includes associations with both conifers and hardwoods, “it gives us a more realistic representation of the Earth’s surface,” says Gopal.

The ART model and its applications exemplify several of CELEST’s (Center for Excellence for Learning in Education, Science, and Technology) primary activities, which include modeling cognition, verifying the models, applying them to real-world problems, and bringing neuroscience into education. These research activities propel the field of brain science forward with both theoretical and practical advancements. As neuroscience research extends into fields like manufacturing, medicine, and beyond, the need for students versed in mind and brain science increases, making CELEST’s educational outreach all the more important.

For more information about CELEST, see http://cns.bu.edu/celest; for ART model applications, see: http://cns.bu.edu/techlab

— Elizabeth Dougherty

This map was produced by applying the ART (Adaptive Resonance Theory) neural model to remote sensing data. It shows changes in land use in the Nile River Delta between 1984 and 1993. ART mimics the way the brain remembers and makes sense of data and images. Its applications are varied — it can be used to streamline a manufacturing process or, as in this case, interpret satellite images to better understand Earth's changing landscape.