At the Beijing Winter Olympics, Gu Ailing had a habit of swinging her body before taking off and "going through" the game in her head. In the view of Wang Liping, a researcher at the Center for Excellence and Innovation in Brain Science and Intelligent Technology of the Chinese Academy of Sciences, if each competition action is regarded as a point in space, Gu Ailing maps these actions to different timings of a sequence and rehearses the memory of the sequence, which will be smoother for the next memory extraction.
The human brain processes sequence information all the time, but we know very little about how the brain "encodes" this sequence information. On February 11, 2022, science, an international authoritative academic journal, published the research results of Wang Liping's research group, Min Bin, associate researcher of Shanghai Brain Science and Brain-like Research Center, and Tang Shiming's research group of Peking University, in the form of a long article. They found that macaque brain neurons characterized every spatial location in the sequence in the form of swarm encodings, and found similar ring geometries in these representations. The study overturns key assumptions of classical sequence working memory models, providing new insights into the conundrum of how neural networks perform symbolic representations that will have an impact on brain-inspired AI.
[There are three independent "screens" in the macaque brain at the same time]
Sequence information is everywhere, and on a larger spatio-temporal scale, the 24 solar terms of the Chinese lunar calendar are also nature's memory of sequence information.
The brain needs to remember the entire sequence before applying timing information. For example, when asking for directions, remember the directions of the guide, and when dancing, remember a series of movements demonstrated by the teacher. Not only do individual pieces of content need to be remembered, but the order between them cannot be confused.
Macaques are evolutionarily the closest model animals to humans, and their cognitive abilities and brain structure and function are closer to humans than other model animals. To explore the problem of timing memory coding, the researchers trained macaques to memorize spatial sequences consisting of multiple location points.
During the mission, three different dots flash in turn on the screen in front of the macaque, and the macaque needs to report the order in which these taps were presented before a few seconds later. During these few seconds of memory retention, spatial sequence information is temporarily stored in the brain in the form of working memory. To record the activity of the brain neuronal population during the macaque monkeys' task, the researchers performed two-photon calcium signal imaging of the lateral prefrontal cortex, the base of working memory. Calcium signaling reflects the pulse firing activity of neurons, and the key to the characterization of sequence information lies in the activity patterns of neuronal populations in these few seconds.
The researchers hypothesized that macaques also had a "screen" in their brains where they could write down the dots that had appeared. How do I represent multiple pieces of information in a sequence at the same time? They further guessed that there were three different "screens" in the macaque's brain at the same time, so that each "screen" only needed to write down the information of one point, and the screens did not interfere with each other.
The high-dimensional data obtained through calcium signal imaging confirmed their hypothesis that the macaque brain did use three separate "screens" to characterize sequence information.
The brain uses three separate "screens" to characterize sequence information
Traditionally, classical sequential memory models have assumed that individual neurons are the basic units of computation, playing similar roles in different orders. This study overturns this hypothesis by finding that a large number of neurons play completely different roles in different orders, suggesting that the "encoding" of sequence memories should be more focused on the population neuron level than on the nature of individual neurons.
[Will have an impact on brain-inspired artificial intelligence]
What is the significance of this original scientific discovery?
Academician Guo Aike, a neuroscientist and biophysicist, commented that the innovation of this work lies in the experimental paradigm of coexistence of two clues of time and spatial information based on the sequence learning of macaque monkeys, revealing the neural coding and characterization mechanism of sequence information on the time scale of working memory.
70 years ago, neuropsychologist Carl Rushri proposed the hypothesis that in order to control sequenced actions, our brains need to instantaneously shift their work to constant patterns of neural activity. "Happily, this finding provides experimental data to support this theoretical hypothesis." That's the charm of science! Guo Aike said.
(From left to right) Wang Liping, Min Bin and the two co-first authors of the paper, Hu Peijue and Xie Yang, discuss
"A core dimension of cognitive activity is time series. The representation and manipulation of time series information is the basis for important cognitive activities including memory and language. How the brain encodes timing information is an extremely important unsolved mystery. Academician Chen Lin, a cognitive science and experimental psychologist, commented that this is a "milestone" type of important work in the field of cognitive neuroscience.
The research will also have an impact on brain-inspired AI. In the field of artificial intelligence in the last century, researchers have proposed that if the representation of symbols by brain neural networks can be imitated, artificial intelligence will have a relatively large leap. "This discovery clearly illustrates the neural mechanisms of sequential working memory and provides new ideas for understanding the conundrum of how neural networks perform symbolic representations." Neurobiologist Academician Wang Yizheng thinks.