Luo Huan and Zhang Hang's research group published a paper in Progress in Neurobiology to reveal the dynamic emergence of relationship network structure in the human brain...

Imagine being watched a drama
.

The curtain opens, the lights come on, and the characters appear
.
As the plot unfolds, we will gradually build a network of character relationships in our minds, such as friends and enemies, couple rivals, bandits A and so on
.
More interestingly, we will further form new knowledge from this, such as reasoning that two people who have never appeared together may be in the same camp
.
It can be said that an important ability of human beings is to constantly pursue the relationship between all things, which is also an important embodiment of
human intelligence.
Therefore, understanding how the human brain learns and reasons about the network structure behind this fragmented information is of great significance
for revealing the basic mechanisms of human intelligence.

In order to study this problem and its brain mechanism, the research group of Professor Huan Luo and Professor Zhang Hang of the School of Psychological and Cognitive Sciences and McGovern Institute of Brain Science of Peking University designed a novel experimental paradigm of sequence prediction and recorded electrical activity
with high temporal resolution.
Fifteen random images (Figure 1A, right) are selected and embedded in a community network structure (Figure 1A, left), and the probability transfer relationship between the pictures is determined by their interrelationship in the network
.
The team recruited human subjects to complete the image prediction task (Figure 1B): they watched the image stream, predicted what the next image would be, and learned
through feedback.
In other words, the subjects had never seen the community network structure, but if they were able to successfully complete the prediction task, it showed that they had learned the relationship between these pictures, that is, formed the abstract community network structure
in their minds.

Figure 1.
Transfer network and sample trials

In terms of behavioral performance, human subjects did learn the "hidden" network relationship structure, which was manifested in the continuous improvement
of predicted performance.
The most core discovery is that the researchers found a neural representation of the abstract relationship network structure in human brain activity, which appeared about 540~930 milliseconds after the picture was presented (Figure 2).

In addition, there is a close relationship between this neural representation and the predicted behavior of the subject, that is, the higher the similarity of the neural representation of the picture, the faster
the prediction is reflected.
Furthermore, the emergence of higher-order statistical structures has been found in behavioral and neural activity, manifested by the properties
of intra-cluster compression and intercluster distant.
Furthermore, by establishing and comparing multiple computational models, the researchers found that the human brain adopts the inheritor representation strategy to learn and reason to form a higher-order structure (Figure 3).

Figure 2.
Neural characterization of low-order metastatic probability

Figure 3.
Emergence of higher-order structures in neural signaling and computational models

In summary, Luo Huan and Zhang Hang's research group combined behavioral, neural and computational models to reveal the neural mechanism and computer theory behind the human brain's extraction and learning from continuous picture streams, and then establishing the neural mechanism and computer theory of high-order statistical structure (intracluster compression and intercluster distance) from this low-order transfer probability
.

The study, titled "Dynamic emergence of relational structure networks in human brains," was published online Nov.
10 in
Progress in Neurobiology, a major journal in cognitive neuroscience.
Dr.
Xiangjuan Ren, a postdoctoral fellow at the School of Psychological and Cognitive Sciences, Peking University, is the first author
of this paper.
Luo Huan and Zhang Hang are co-corresponding authors
of this article.
This research was supported
by the National Science and Technology Innovation 2030 Major Project, the National Natural Science Foundation of China Key Project and General Project, Peking University New Engineering Program, Peking University-Tsinghua Joint Center for Life Sciences, and Peking University-Tsinghua Joint Center for Life Sciences Postdoctoral Program.