• New research from MIT neuroscientists suggests how the brain learns which category an object belongs to — for example, fruits or animals.

    New research from MIT neuroscientists suggests how the brain learns which category an object belongs to — for example, fruits or animals.

    Image: Christine Daniloff

    Full Screen

How the brain assigns objects to categories

New research from MIT neuroscientists suggests how the brain learns which category an object belongs to — for example, fruits or animals.

New findings may explain why children with autism tend to fixate on details instead of seeing the big picture.

The human brain is adept at recognizing similar items and placing them into categories — for example, dog versus cat, or chair versus table. In a new study, MIT neuroscientists have identified the brain activity that appears to control this skill.

The findings, published in the July 27 issue of the journal Neuron, suggest a potential explanation for why autistic children focus intently on details, but often seem unable to group things into broad categories, says Earl Miller, the Picower Professor of Neuroscience and senior author of the paper.

"We think what may happen in autism is the system may get out of balance … and as a result, the details overwhelm the category. Then you have a brain that's not only too good at memorizing details, it can't help but memorize the details," says Miller, a principal investigator at the Picower Institute for Learning and Memory at MIT.

Miller and Picower postdoc Evan Antzoulatos focused their study on two brain regions, the prefrontal cortex and the striatum, which is part of a larger structure known as the basal ganglia. Both regions are known to be important for learning.

Until a few years ago, it was believed that the prefrontal cortex learns information quickly, then sends what it learns to the basal ganglia, which helps form habits, such as the ability to play a musical instrument. However, in 2005, Miller and colleagues showed that when monkeys learn simple tasks, their basal ganglia are more active early in the process, followed by a slower activation in the prefrontal cortex.

In other words, the striatum quickly learns the individual puzzle pieces, and the prefrontal cortex puts them together, Miller says. He and Antzoulatos theorized that the same pattern would be evident during category learning.

For the new Neuron study, Antzoulatos trained monkeys to assign patterns of dots into one of two categories. At first, the animals would see only two examples, or "exemplars," from each category — a small enough number that they could memorize the category to which each belonged, without having to learn the general category traits. After the animals learned the first two exemplars, the number would be doubled. Eventually, the number of exemplars became so great that it was impossible to memorize them, and the monkeys' brains would start picking up on general traits that characterize each category.

As they did so, brain activity shifted from the striatum, a more primitive brain region, to the prefrontal cortex, which is responsible for high-level functions such as planning and decision making.

"What happens during category learning is the more primitive, faster basal ganglia can memorize the exemplars, but then it sends what it learns up to the prefrontal cortex. And the prefrontal cortex figures out what's common among all the exemplars, among all the individuals, and extracts the essence," Miller says.

Gregory Ashby, a professor of psychology at the University of California at Santa Barbara, says the new study represents the "clearest picture yet" of the striatum's involvement in category learning. "We've known for quite a while that the striatum plays an important role in category learning, but it was not at all clear exactly what that role was," he says.

In future studies, the MIT researchers hope to test their theory that autism results from an imbalance between the striatum and prefrontal cortex by interfering with the normal balance between the two brain regions and observing the results.

Topics: Autism, Brain and cognitive sciences, Learning, Memory, Picower Institute for Learning and Memory


I wonder if the reverse is true. If you are especially good at sorting objects, are you likely to have trouble remembering details?

Is it possible that humans could categorize their pain and in order to know if it is a serious thing that could affect their lives? Besides, thinking about Cheryl's question, how would work the recall of a pain experience?

I believe we have missed a fundamental characteristic of the axon by a physical oversimplification. If we model the axon as Colinear Coaxial Cable instead of a linear cable things get interesting. The justification is simple. We know that there is a distribution of ions that form on charged surfaces, the electric double layer (EDL) which falls of exponentially with distance. The myelin sheath is a dielectric separating the cytoplasm from the extracellular fluid. It then follows that the EDL behaves as the outer conductor whereas the cytoplasm behaves as the inner conductor separated by the dielectric. At the Nodes of Ranvier, where there is no myelin the extracellular fluid and cytoplasm are separated by the axonal membrane. Since there is active diffusion of ions across the membrane the inner and outer conductors are effectively connected to each other, thus a colinear coaxial cable. So what? This cable radiates. What if neurons broadcast their membrane potentials in the far-field? It would be analogous to having a ham radio and trying to find your friend by cycling through different frequencies. Eventually you would find each other.

Now consider that there are committees of neurons (or neocortical columns) that tune in on certain frequencies to talk to each other. Now this is exciting because what we have is a graph with colored edges where the objective is to get as many monochromatic complete subgraphs as possible. Ramsey's Theorem! For any sufficiently large complete graph where we color the edges there will always exist monochromatic complete subgraphs! In other words if neurons are radiating at random frequencies where every other neuron can detect then by the shear number of neurons there will always be order out of chaos. Regions of the brain MUST synchronize! But wait there is more! These combinatorially inevitable monochromatic complete subgraphs behave as Hopfield Networks!

Man I want to go to MIT's Brain and Cognitive Science graduate program and test this hypothesis. Man this stuff is cool!

Back to the top