[ad_1]
This use of orthogonal encoding to separate and protect information in the brain has been seen before. For example, when monkeys are preparing to move, their neural activity in the motor cortex indicates potential movement. it does so orthogonally so as not to obstruct with signs that carry real orders to the muscles.
However, it has often not been clear how neuronal activity is transformed in this way. Buschman and Libby wanted to answer that question because of what they were observing in the hearing cortex of their mice. “When I first started in the lab, it was hard for me to imagine how something like this could happen with shooting activities,” Libby said. “He wanted to open the black box of what the neural network does to create that orthogonality.”
Examining the options experimentally, different subsets of neurons in the auditory cortex ruled out the possibility of manipulating sensory and memory representations independently. Instead, they showed that the general population of neurons was the same and that the activity of neurons could be neatly divided into two categories. Some had “stable” behaviors in representations of the senses and memory, and others that “moderately” changed neurons reversed response patterns for each use.
To the astonishment of the researchers, this combination of stable and continuous neurons was sufficient to rotate sensory information and convert it into memory. “That’s all the magic,” Buschman said.
In fact, he and Libby used approaches to computational modeling to show that this mechanism was the most effective way to construct orthogonal representations of sensation and memory: it required fewer neurons and less energy than alternatives.
The findings of Buschman and Libby fuel a new trend in neuroscience: populations of neurons, even in the lower sensory regions, engage in richer dynamic coding than previously thought. “These sections of the lower cortex in the food chain have some very interesting dynamics that we may not have greatly appreciated so far,” he said. Miguel Maravall, A neuroscientist at the University of Sussex who was not involved in the new research.
The work can help to reconcile the two sides of the debate through short-term memories through continuous and sustained representations, or through dynamic neural codes that change over time. Instead of going down one side or the other, “our results show that basically both were right,” Buschman said, getting the first stable neurons and the second changing the neuron. The combination of processes is useful because it “helps prevent interference and make that orthogonal rotation”.
The research of Buschman and Libby may be important in contexts that go beyond sensory representation. They and other researchers hope to look for this mechanism of orthogonal rotation in other processes: to find out how the brain monitors multiple thoughts or goals; how he deals with a task while dealing with distractions; how it represents internal situations; how it controls cognition, including attention processes.
“I’m very happy,” Buschman said. Looking at the work of other researchers, “I remember seeing that there is a stable neuron, a changing neuron! You see them everywhere now. “
Libby is interested in the implications of the results in artificial intelligence research, especially in the design of architectures that are useful for AI networks that have multiple roles to play. “I would like to see people pre-assign neurons in neural networks to have stable properties and change properties, rather than random properties, to help their networks in some way,” he said.
Ultimately, “the implications of this information coding will be very important and very interesting to know,” Maravall said.
Original story reprint with permission Quanta magazine, independent publisher’s publication Simons Foundation its mission is to improve public knowledge of science by covering developments and trends in research in mathematics and the physical and physical sciences of life.
More great KABEKO stories
[ad_2]
Source link
