A new research approach makes it possible to document how the brain learns new rules
Learning rules is essential to the survival of many animals. It allows them to identify recurring patterns in their environment and derive rules from them in order to deal with similar situations in the future. The ability of primates, and in particular humans, to learn rules is particularly developed and enables complex decision-making and long-term planning. Since the natural environment is rich in information and constantly changing, the main challenge facing the brain in the learning process is to distill from the surrounding "noise" valid laws that will be relevant even in situations different from those in which they were learned.
Right: Prof. Roni Paz and Prof. Elad Schneidman, Weizmann Institute. The game person
While these learning processes have been extensively studied at the behavioral level, real-time monitoring of the learning process as it occurs inside the brain is complicated. New research של Prof. Roni Paz, Prof. Elad Schneidman And Dr. Yordan Cohen from the Department of Neurobiology at the Weizmann Institute of Science, is breaking new ground in the field by combining computational models and recording the activity of nerve cells, from the very first moments of the learning process.
The institute's scientists developed a learning task that requires sorting small patterns, each of which has several black or white squares in a different order. In each round of the experiment, the researchers define a rule that divides the templates into two groups. For example, one group where in all patterns the right square is black - and another group where it is white. The learning process in each round is one of trial and error. Each time the subject is presented with one pattern, and he has to decide which of the two groups to associate it with. In a successful round, the subject gradually learns the law defined by the researchers according to the feedback he receives.
This learning task turned out to be very effective: the game pieces - the patterns - are relatively simple, but a very large variety of rules - both simple and complicated - can be created from them. These patterns have another and important advantage: all the laws, that is, all the possibilities of dividing the patterns into two groups, can be represented in a geometric space that enables the quantification and processing of the data to extract meaningful information from them. The spatial representation is done with the help of a multidimensional cube, and each law is represented as a vector - an arrow in a certain direction inside the cube.
During the experiment, the researchers recorded the activity of nerve cells from two areas of the brain, which are known to be important for learning rules: the frontal cortex and the basal ganglia. The researchers measured and quantified the response of each neuron in relation to each presented pattern. At this point lies an important achievement of the research: the researchers processed the information on the activity of the nerve cells so that it is also represented with the help of a vector located in the same multidimensional cube where the studied laws are represented.
The neurons in the cerebral cortex "learn" through a process of trial and error
The fact that the brain activity and the learned law are represented in the same space, made it possible to reflect the learning at the nerve cell level. Since each law is represented by a specific vector, the activity of the single nerve cell is reflected by a series of vectors that each time stand in a different place in space and "search" for the correct law. That is, in a situation where learning is progressing, the vectors of the neural activity approach the place where the vector that represents the law is located.
In this way, the researchers found that the nerve cells in the frontal cortex "learn" through a process of trial and error, as they get closer and closer to the law. In contrast, the vectors of the nerve cells in the basal ganglia are lengthened - that is, they establish the brain's confidence that it has found the right law.
The researchers also discovered that there is coordination between the processes taking place in the different parts of the brain: the basal ganglia strengthen the brain's confidence in the law immediately after the neurons in the frontal cortex recognize it. Furthermore, the neural activity at the end of an experimental day made it possible to predict with high probability the behavior and the success in performing the task the next day. This fact strengthens the researchers' confidence in the correct identification of the areas of the brain where the learning process takes place, as well as the reliability of the model they developed. In the future, this model will be able to allow a relatively accurate measurement of various disturbances to the learning process and the planning of directed learning that will improve and accelerate learning ability.
