This use of orthogonal coding to separate and protect information in the brain has already been seen. For example, when monkeys prepare to move, neural activity in their motor cortex represents potential movement but does it orthogonally to avoid interfering with signals leading real commands to the muscles.
Yet it is often difficult to know how neural activity changes in this way. Buschman and Libby wanted to answer this question for what they observed in the auditory cortex of their mice. “When I first started in the lab, it was hard for me to imagine how anything like this could happen with neural triggering activity,” Libby said. She wanted to “open the black box of what the neural network does to create this orthogonality.”
By experimentally examining the possibilities, they ruled out the possibility that different subsets of neurons in the auditory cortex independently handle sensory and memory representations. Instead, they showed that the same general population of neurons was involved and that the activity of neurons could be clearly divided into two categories. Some were “stable” in their behavior during sensory and memory representations, while other “switch” neurons reversed the patterns of their responses with each use.
To the researchers’ surprise, this combination of stable neurons and switching was enough to rotate sensory information and turn it into memory. “That’s all the magic,” Buschman said.
In fact, he and Libby used computer modeling approaches to show that this mechanism was the most efficient way to build orthogonal representations of sensation and memory: it required fewer neurons and less energy than the alternatives. .
The findings of Buschman and Libby are fueling an emerging trend in neuroscience: that populations of neurons, even in lower sensory regions, are engaged in richer dynamic coding than previously thought. “These parts of the cortex that are further down the food chain also have some really interesting dynamics that we maybe haven’t really enjoyed until now,” said Miguel maravall, a neuroscientist from the University of Sussex who was not involved in the new study.
The work could help reconcile both sides of an ongoing debate over whether short-term memories are maintained by constant and persistent representations or by dynamic neural codes that change over time. Instead of going down one side or the other, “our results show that basically they were both right,” Buschman said, with stable neurons doing the first one and switching neurons on the second. The combination of processes is useful because “it actually helps prevent interference and makes this orthogonal rotation.”
Buschman and Libby’s study may be relevant in contexts beyond sensory representation. They and other researchers hope to find this mechanism of orthogonal rotation in other processes: in the way the brain keeps track of several thoughts or goals at once; in the way he engages in a task while dealing with distractions; in the way it represents internal states; in the way it controls cognition, including attentional processes.
“I’m really excited,” Buschman said. Looking at the work of other researchers, “I just remember seeing, there is a stable neuron, there is a switching neuron!” You see them everywhere now. “
Libby is interested in the implications of their findings for artificial intelligence research, particularly in the design of useful architectures for AI networks that must multitask. “I would like to see if the people who pre-allocate neurons in their neural networks to have stable and switching properties, instead of just random properties, have helped their networks in some way or another,” she declared.
Overall, “the consequences of this kind of information coding are going to be really big and really interesting to understand,” Maravall said.
Original story reprinted with permission from Quanta Magazine, an editorially independent publication Simons Foundation whose mission is to improve public understanding of science by covering developments and research trends in mathematics and the physical and life sciences.
More WIRED stories