The brain clings to old habits when learning new tricks

Original author: John Rennie
  • Transfer

The adaptability of the brain sometimes seems endless. But the observation of the brain in the learning process suggests that its neural networks can be surprisingly inflexible and inefficient





The main sign of intelligence - the ability to learn. As decades of research have shown, our brain shows a rather large "plasticity", that is, the ability of neurons to change connections in response to external stimuli. But researchers at Carnegie Melon University and the University of Pittsburgh recently discovered unexpected limitations on our learning abilities. The brain can in fact be flexible and adaptive, but at least for short periods of time, it learns through the ineffective repetition of tricks available in its neural repertoire, rather than creating a network of connections from scratch.

“When I play squash , it’s like a tennis player,” said Byron U, a biomedical engineer and neurobiologist at Carnegie Mellon University, as well as one of the leaders of the study. Yu played tennis for many years. He has problems with squash because he uses shorter rackets and makes faster and more difficult punches - unlike those used on a tennis court. However, when playing squash, he rolls over to the style of using rackets, which he has been taught by the long experience of playing tennis. The brain is not so easy to part with what he already knows.

Observing the work of the brain during training, Yu and colleagues found evidence of a similar lack of plasticity at the level of neurons. This discovery and related research can help explain why some things are harder to learn than others.

A few years ago Yu, Aaron Batistafrom the University of Pittsburgh and members of their laboratories began to use the brain-computer interfaces (IMC) as tools to make discoveries in neuroscience. These devices have a chip the size of a nail, capable of tracking the electrical activity of a hundred neurons right away in the motor cortex of the brain that controls movements. By tracking the sequence of voltage spikes coming through individual neurons, the IMC is able to calculate the “burst speed” characterizing the behavior of each neuron during the execution of a specific task.

“You can imagine the difficulties associated with digging into the heap of this data in an attempt to determine what the brain is doing,” Y. said. “Our eyes are not very trained to notice hidden patterns in such data.” But the advanced statistical analysis that the chip is capable of can do this, and these patterns can be used to determine the nervous activity associated with the experimental intent to make certain movements. For example, the system is able to discern the subject's intentions to reach out with his hand left, right, up or down.

Researchers can then use the IMC output to translate the neural activity associated with a particular movement into commands for the cursor on a computer screen. By trial and error, people or animals are trained to use such an interface, imagining how they move their hand, say, to the left, and moving the cursor in the same direction.

When Yu, Batista and their colleagues tracked the monkey's motor cortex, which again and again performed simple hand movements, they found that the neurons did not activate independently. The behavior of the measured hundreds of neurons could be described statistically through the behavior of 10 neurons, which in various ways activated or suppressed the work of their neighbors. In the analysis, this result was shown as a set of points that filled a very small amount in a 100-dimensional data space.

“We called this volume a characteristic set [intrinsic manifold], because we believe that this feature is truly characteristic of the brain,” said Stephen Chase , a professor of biomedical engineering at Carnegie Malone University. “The dimension of this space accurately predicts the capabilities of neurons.”

In 2014, researchers discoveredthat it is easier for subjects to study new tasks if combinations of neurons belonging to a characteristic set, rather than those that lie outside it, are included in this process. According to Yu, this makes sense, since tasks that fall into a characteristic set make inquiries to the brain corresponding to the underlying nervous structure. Upon completion of the study, the group switched to the question of how nervous activity changes during training — this is described in a paper recently published in the journal Nature Neuroscience.

To understand what is going on in the brain, the researchers first gave primates equipped with an IMC to get comfortable with moving the cursor left and right. The team then changed the requirements for what kind of nervous activity is needed to move the cursor, and began to observe which new patterns of nervous activity corresponding to the new set of points of the characteristic set will be used by animals.

The researchers expected to see evidence of a learning strategy called “restructuring,” during which the animal would begin to use some kind of new neuron circuitry that would most naturally fit the task. “Perestroika is the best strategy that animals can use, subject to the limitations of a characteristic set of neurons,” said Matthew Golub, postdoc, who participated in this project with Yu and Chase, and is currently working at Stanford University. Or the monkey brain could start learning through “rescaling”, a process in which the neurons involved in the primary learning task would increase or decrease the number of bursts until they stumble upon some kind of work flow.

But, to the great surprise of the researchers, neither perestroika, nor rescaling happened. Instead, researchers observed a highly inefficient “reassociation” process. The subjects were trained in a new task, simply using the existing nerve sequences, changing their purpose. The sequences that previously moved the cursor to the left began to move it to the right, and vice versa. "They were engaged in reuse," said Golub, only under new conditions.

Why use the brain is not the most effective learning strategy? The results of the discoveries of the group suggest that as the work of the entire nervous architecture is limited by the activity of the characteristic set, so the work of neurons from this set has limitations on the reorganization of their activity. Batista suggests that changes in synaptic connections between neurons that would have had to be carried out for restructuring would be too complicated to be carried out quickly enough. “The ductility in the short term may be more limited than we assumed,” he said. - Learning implies forgetting. The brain reluctantly refuses the acquired skills with which he already knows how to handle. ”

Chase compared the motor cortex with an old telephone switchboard, in which nerve connections, like cables, connect the inputs of cortex sections to the cerebellum outputs. As he said, during the experiments “the brain just changes the cable connection pattern” - although the nuances of this process are still unclear.

“The strategy of quick change implies a change in the input connections of the cortex,” said Y. But he also noted that in their experiments, brain activity was monitored for 1-2 hours. Researchers cannot yet exclude the possibility that reassociation serves as an intermediate stage in teaching the brain new tasks; for a longer time, rebuilding or rescaling may still occur.

If so, this may explain the difference in how new information related to common interests is processed by beginners and experts. “Beginners have to work with what they have, and experts are engaged in the consolidation of knowledge,” said Batista. “The described may be the neural basis for this well-known phenomenon.”

Also popular now: