Postdoc Spotlight: Adaptive Learning

Rachel St. Clair, Ph.D., is a postdoctoral fellow in the FAU Center for the Future Mind, and one of the first to earn a doctorate degree in computational neuroscience at the FAU Stiles-Nicholson Brain Institute.

She was recently awarded a grant from Cisco Systems for her research titled Using High-Dimensional Computing for Knowledge Graphs Towards Efficient General Intelligence. Her research will focus on how “true artificial intelligence, or what is sometimes referred to as artificial general intelligence, will likely require the ability of computer programs to do adaptive learning.”

Some key markers of adaptive learning are the ability to learn new information without forgetting prior learning, St. Clair said. While this task may sound trivial, it is often hard for programs to do so without demanding excessive use of computational resources, a significantly different mode of learning than is observed in biological organisms. Being able to learn within a resource constrained environment and to extend that learning to data not seen before quickly and accurately are important first steps in generating the desired program.

A popular model for learning in projects that seek artificial general intelligence are knowledge graphs. These programs make use of a graph-like structure such as edges and vertices to organize information into meaningful representations which can later be called upon during information retrieval. As any graph model is expected to learn continuously, the graph grows exponentially, slowing the time to act and consuming more resources. There is also the trouble of getting the information to be stored in the graph program in such a way that makes sense over a variety of types of knowledge without causing an explosion of memory. This is where abstract, or symbolic representations can be used, St. Clair said.

“We propose that by using a type of math called hyperdimensional computing, knowledge graphs can become more resource efficient, and it is expected that the graphs are more accurate in adaptive learning tasks,” she said.