Carnegie Mellon University

bottleneck traffic

February 18, 2019

Information bottlenecks between brain areas

By Emily Durham

Krista Burns

Most brain functions—including how we sense, how we reason, and how we act—rely on interactions between brain areas. However, we are only just beginning to understand how neurons in different parts of the brain mediate these interactions.

The brain is able to sort which kind of information is passing from one area of the brain to another, according to a paper published today inNeuron, one of the premier journals in neuroscience. This paper is among the first published works looking at how populations of neurons interact between brain areas. 

The research team—which included Carnegie Mellon University’s João Semedo, an ECE postdoctoral researcher, and Byron Yu, an associate professor of biomedical engineering (BME) and electrical & computer engineering (ECE), as well as Adam Kohn of the Albert Einstein College of Medicine and Christian Machens of Champalimaud Research—found that communication between brain areas occurs through an information bottleneck, which they’ve termed a “communication subspace.” Certain types of information are able to “fit” through this bottleneck and others are not.

“Imagine a toddler’s shape sorting game,” says Semedo, who is also a 2018 graduate of the CMU|Portugal ECE Ph.D. program. “The star goes into the star-shaped hole and the triangle goes into the triangle-shaped hole. It’s all about the right information going into the right area—a square peg can’t fit into a round hole, after all.”

Observing two areas of the visual cortex, V1 and V2, Semedo and his team saw that certain population activity patterns were being selectively routed between the two brain areas. This “communication subspace” could be a mechanism by which different types of information, such as color and motion, encoded by one brain area are selectively routed to the appropriate downstream brain areas.

“Communication subspaces might be a general mechanism by which brain areas interact with each other,” says Yu. “So far, this concept has only been studied in the visual and motor cortices. There is still much to be explored.”

For one area to communicate only certain information to another area while keeping other information “private,” in other words, could be an evolutionary benefit that allows a brain area to perform some computation, and then convey only the result of that computation to the appropriate downstream brain area. In other words, the downstream area need not see every detail of the upstream computation.

Understanding these communication subspaces could have many implications for understanding how the coordination between brain areas gives rise to brain function. By better aligning the activity of populations of neurons with their designated communication subspaces, more information can be passed from one area of the brain to another, allowing for better attention, better learning, and better execution of tasks. This paper will enable much further research to be done in these areas.