Mind moving matter

Penn State engineering student Alayna Kennedy is exploring neural networks that will enable more user-friendly prosthetic devices.

Schreyer Honors College Scholar Alayna Kennedy used a research opportunity to find a more effective neural network that will help connect people with prosthetic devices. Credit: Pat LittleAll Rights Reserved.

The human brain is the most complex and sophisticated processor in the world. And some of those brains, including the one belonging to Penn State junior engineering science major Alayna Kennedy, are training machines to learn as humans do.

Kennedy, a Schreyer Scholar, spent the summer researching the applications of machine learning to prosthetic devices, and she will present a paper she co-authored with a University of Colorado-Colorado Springs (UCCS) professor at the second International Conference for the Advancement of Bioscience and Bioengineering from Oct. 26-28 in San Francisco, California.

As part of a Machine Learning Research Experience for Undergraduates (REU), Kennedy spent two months essentially training a neural network to predict body movement via analysis of electromyogram (EMG) signals, which help prosthetic devices understand and replicate muscle movements.

“I took all these EMG signals from different muscles and I created a network to predict how fast someone is walking,” Kennedy said. “I was trying to find the best neural network structure to best predict these different types of walking speeds.”

After testing her structure on a variety of network architectures, she wrote a paper with Rory Lewis, an assistant professor in UCCS’ Department of Computer Science Computer Engineering and Applied Science. Lewis’s other work has examined using neural networks to quickly and accurately diagnose injured soldiers in trauma situations using a small brain implant.

Kennedy has also talked to Vasant Honovar, the director of Penn State’s Artificial Intelligence Research Laboratory, about working on some biologically inspired algorithms. One of her next potential projects would also involve prosthetics. 

“If you have a mechanical hand that you attach to someone’s arm, the device has to be trained,” she said. “This work is basically training the device based on the person’s EMG signals and neural signals so that device is more effective, but what also needs to be done is the person’s brain cells actually need to be rewired. They have to train themselves to use this device, and the more training they put in, they get better and better at using these devices.

“We were thinking about coming up with some algorithm to help people retrain and adapt to this device. That’s still in its very first phases. I’m also hoping to turn this kind of topic into my honors thesis.”

As part of the REU program, Kennedy and the other students had to read a textbook on machine learning and watch several hours of videos prior to when they actually set foot on the UCCS campus in June. (The coding classes she had taken at Penn State helped, too.) She got to know a little bit about other aspects of machine learning, including natural language processing and self-correcting code, by observing the work of those students throughout the summer.

Those experiences piqued what was an already strong interest in machine learning for Kennedy, who liked how various disciplines — coding, mechanical engineering and neuroscience — come together in the field of prosthetics. Her research could help advance it even further and potentially narrow the currently pricey gap between academic and industry prosthetics.

“I just think it’s so cool; prosthetic devices could end disability — within our lifetime,” she said. “Physical disability could be not a thing anymore.”

Last Updated September 16, 2016