When you try to solve a math problem in your head or remember the things on your grocery list, you’re engaging in a complex neural balancing act — a process that, according to a new study by Brown researchers, isn’t about your memory’s storage capacity, but how efficiently it’s being used.
Researchers created a computer model that simulates the interaction between the thalamus — which assists with short-term memory — and the basal ganglia in the human brain, revealing the strategies that our brain employs to manage and encode sensory information. Researchers on the study told The Herald that despite the theoretical nature of the study, the model could help scientists learn more about neurodegenerative diseases and other disorders.
Researchers compared two models, each containing “stripes” that represented separate storage units. One contained eight stripes, while the other used a streamlined “chunking” method — which merged similar information into an averaged representation — with just two stripes. The study found that the “chunk model was able to efficiently use its stripes and actually outperform the model with eight stripes,” according to Aneri Soni PhD’23, lead author of the study.
Chunking could reduce errors when the brain is overloaded with information, according to Soni.
Even though “we have a lot of neurons and … the ability to potentially represent many items,” the challenge is in avoiding errors that arise when too many items are present, Soni explained.
The model must not only store information, but also decide how to allocate its limited resources to place data and merge similar items. Sometimes, when the model was tasked with remembering four or five items using only two stripes, it learned “to use one stripe, but kind of just left the other stripe empty,” Soni said.
These decisions dig into the underlying “management problem” of working memory, a challenge that, as Soni notes, sets the stage for further exploration into how these limitations might be addressed in future research.
For her experiment, Soni used the color wheel task, a popular tool in visual working memory studies. In this task, the model is presented with colored bars of varying orientations and later prompted to recall the color associated with particular orientations.
The model “has to correctly associate the color and the orientation of a bar together, and it has to maintain that information over some delay,” Soni said. The task allowed her to compare the model’s performance with human behavior data collected by the lab in a previous study.
Inspiration for the model traces back to earlier research by Matt Nassar, an assistant professor of neuroscience and cognitive and psychological sciences, which showed how chunking can work in a computational model but didn’t explore how this could play out in humans.
Michael Frank, director of the Carney Center for Computational Brain Science and principal investigator of the paper, explained that the impact of the model extends far beyond theoretical neuroscience and could shed insight to working memory deficits seen in diseases like Parkinson’s.
These disorders could disrupt this new “chunking” process detailed in the paper leading to difficulties in filtering out irrelevant information, he said.
Frank added that the basal ganglia, a critical structure involved in regulating cognitive control, acts as a gatekeeper for information entering working memory.
It prevents “you from accessing more than just a few items at a time by … forcing the system to only focus on a couple of different representations at once,” he explained. The model shows that when this gating process fails, the system becomes overwhelmed by competing signals, making it difficult to manipulate information effectively.
Edward Awh, a professor of psychology and neuroscience at the University of Chicago, who has spent over thirty years studying human memory, explained that understanding the neural processes that govern working memory is essential because “the contents of working memory … define 100% of your lived existence.”
Despite having an abundance of neurons, he explained, humans are typically limited to holding only three to four items in mind.
“If we want to understand working memory capacity limits, then the ultimate accomplishment would be to build a concrete … neural model of how information is encoded into working memory,” he said, adding that he is “excited” about the model’s focus on efficient resource management.
Awh, who was not involved in the Brown study, noted the clinical relevance of these findings for disorders such as ADHD and schizophrenia. He believes that a concrete model of working memory could guide targeted interventions to help restore cognitive function in affected individuals.
“Anything we can do to learn about the systems that are challenged is going to help us to conceive of new interventions,” he said.
Correction: A previous version of this article misstated quotes from Aneri Soni, Michael Frank and Edward Awh. The article has been updated to accurately reflect Soni’s statements on neuronal capabilities, the behavior of the two-stripe model and the model’s responsibilities in the color wheel task. Changes have also been implemented to accurately reflect Frank’s sentiments on filtering information and the basal ganglia. The article has also been updated to accurately reflect Awh’s comments on the contents of working memory and the future of working memory models. The Herald regrets these errors.
