Mechanisms of skill learning
Most people can swing a hammer, but they can’t swing it with the speed and precision of a master carpenter. The difference is thousands of hours of practice and the systematic organization of hundreds of thousands of the brain’s neurons.
Practice makes perfect, in a pretty literal sense. When we improve at a skill over time, it is presumably driven by coordinated changes in our brain’s neural representation of how that movement should be completed,” explains Steven Chase, an associate professor of BME. “Yet, the link between how our brain reorganizes its neurons and how we learn a new skill is still largely unknown.”
Chase was awarded a five-year, $800,000 NSF CAREER Award to discover the link between the neural reorganization and skill learning. With the award, he will research the behavioral factors that drive skill learning.
Chase is a member of CMU’s BrainHub, which builds on its strengths in biology, computer science, psychology, statistics, and engineering.
An improved understanding of the science behind skill learning will have long-term impact on the clinical understanding of the progression of various motor control disorders, such as Parkinson’s disease and stroke. His research may inform the design of targeted rehabilitation programs for those patient groups.
“You can imagine stroke as a sort of rewiring of the brain’s system. Because parts of the brain are now dead, there are neurons that contribute completely differently to that circuit,” says Chase, who also is a member of the Center for Neural Basis of Cognition, a joint project between Carnegie Mellon and the University of Pittsburgh devoted to investigating the neural mechanisms that give rise to human cognitive abilities. “In stroke rehabilitation, the brain must learn to use those neurons in an appropriate way for this altered system. We want to understand how the brain does this learning.”
A major challenge in studying skill learning is that most movements engage tens of thousands of neurons, and the link between any individual neuron and movement is not known. To overcome this problem, Chase and his lab will use a brain-computer interface, a device that allows the brain to control a computer cursor using thought alone, and observe how neurons change when mastering control of the device. By using a brain-computer interface, Chase says the group can interpret how changes in individual neurons combine to enable skill development.
“Sometimes our brain actually requires us to rebuild a neural circuit in order to make what was previously impossible, possible,” Chase says. “With this award, we will go deep into that process and answer the question: ‘how do you rebuild those neural circuits?’”
Studying light to advance computer vision and more
Aswin Sankaranarayanan, assistant professor of ECE, received an NSF CAREER Award of approximately $532,000 to study how light interacts with materials using light rays and their transformations.
The project, titled “Plenoptic Signal Processing—A Framework for Sampling, Detection, and Estimation Using Plenoptic Functions,” will explore how light interacts with objects in a scene by studying characterizations of light that go beyond images. Specifically, Sankaranarayanan will study how light varies with angle and spectrum, in addition to space and time, and how these attributes of light change after interacting with a scene.
“With this award, my team will develop imaging systems and associated algorithms for sensing and interpreting interactions of light with materials. The visual complexity of real-world scenes implies that these interactions are often intricate and complex,” says Sankaranarayanan. “Our goal is to study these interactions at unprecedented space and time resolutions, thereby advancing research in many disciplines including computer vision, graphics as well as 3-D acquisition and printing.”
One goal of the project is to develop robust 3-D scanning of objects, which will enable vision systems to deal with a wider range of objects and imaging conditions. Sankaranarayanan’s team will use this research to build computational cameras that are better equipped to handle specific tasks than traditional cameras.
Infrastructure management under uncertainty
Matteo Pozzi sees great potential in sensors and robotic technology that collect data to help inform decision-making related to infrastructure. The NSF sees great potential in Pozzi, and has given the CEE associate professor a $500,000 CAREER Award to suggest strategies that are optimal for collecting information and for taking actions. Through integrating models and computational approaches, Pozzi hopes to optimize infrastructure operation and maintenance, and the continued collection of information.
“Because we are managing such limited resources, data collection and this process of learning about the infrastructure must be optimized,” he says, proposing that algorithms could offer guidance on where and when to add more sensors, schedule inspections, or conduct strategic testing.
“Managers have to compare the benefits of collecting information with the benefits of repairing various components, where each choice is expensive,” he says. As Pozzi establishes and refines his algorithms, he also will develop methods to teach infrastructure planning and analysis. Partnering with CMU’s Summer Engineering Experience for Girls program, Pozzi plans to build a simulation game in which students act as virtual infrastructure managers who must develop, test, and revise decision-making strategies in the face of persistent risk and uncertainty.
“I’m excited because it’s an expansive, long-term project that allows me to investigate topics I am passionate about, to educate students and to form a path in the direction in which I want to research and teach,” he says.