Reinforcement learning — a process during which humans alter their behavior as they learn which behaviors yield the best results — and working memory — which allows people to optimize decision-making by keeping track of previous actions and their consequences — appear to be two separate modes of human learning. But a recent study has demonstrated that the two processes cooperate to assist humans as they learn.
Anne Collins, assistant professor in the department of psychology at the University of California at Berkeley and former postdoctoral researcher at the University, and Michael Frank, professor of cognitive, linguistic and psychological sciences, co-authored the study.
“People used to view these two systems as independent of each other during the learning process,” Frank said. “This study shows that reinforcement learning and working memory work together when they operate, rather than compete.”
Participants in the study had to learn associations between symbols and corresponding buttons on a keyboard, according to a University press release. Researchers recorded brain activity during the learning process using EEG, which provided real-time data, Frank said.
“We varied the number of items shown on the screen,” ranging from “numbers less than the average capacity working memory and numbers more than the average capacity,” Frank said. There is a limited amount of items that can be held in working memory, and the duration of time which these items can be held is also finite. Taken together, when the number of symbols and the delay between the symbols increase, the working memory component in the learning process should shrink, according to the press release.
The study revealed that brain signals associated with reinforcement learning were stronger when participants were given stimuli that overloaded working memory. But these signals were weaker when participants had to hold fewer items in working memory, Frank said. These findings suggest that the two systems are dependent on each other, because varying the inputs for working memory should not alter reinforcement learning signals if they were independent systems.
“While this study examines learning at a more basic science level of how discrete learning systems work in the brain, there are probably learning applications in the long run, but we are not at that stage yet,” Frank said. “However, understanding the biology of the two different systems can allow us to better predict how learning processes are affected among people with mental disorders and illnesses that affect these parts of the brain.”
“This is quite an important finding that advances our understanding of the learning and memory systems that drive adaptive behavior,” said Candace Raio, a postdoctoral fellow at the Center for Neural Science at New York University. “Knowing that these reinforcement learning and working memory systems work cooperatively constitutes a major step forward in understanding how we learn and adapt to environmental demands.”