Contrary to previous assumptions, the wiring of nerve cells in the human neocortex differs from that of mice. A study found that human neurons communicate in one direction, while in mice, signals tend to flow in loops. This distinction enhances the efficiency and capacity of the human brain in processing information. The discoveries from this research could also advance the development of artificial neural networks. These findings were published in the journal Science after being conducted by Charité — Universitätsmedizin Berlin.The research revealed that in humans, neurons communicate in a single direction, while in mice, signals tend to form loops. This difference in neural communication enhances the efficiency and capacity of the human brain to handle information. These findings could have implications for the advancement of artificial neural networks.
The neocortex, a crucial component of human intelligence, is less than five millimeters thick. This region of the brain houses 20 billion neurons that are responsible for processing sensory input, planning actions, and contributing to consciousness. The mechanism by which these neurons manage such complex information is largely dependent on the neocortex of animals such as mice, which allows for different information processing. Professor Jörg Geiger, Director of the Institute for Neurophysiology at Charité, explains that our previous understanding of neural architecture in the neocortex was primarily based on findings from animal models. In these models, neighboring neurons communicate with each other in a dialogue-like manner, often sending signals back and forth in recurrent loops. However, the human neocortex is thicker and more complex, allowing for a different and more intricate processing of information.The connectivity of the human brain is complex and not easily understood, but researchers previously believed it to be similar to that of a mouse. However, due to a lack of data, this assumption was never confirmed. A team of researchers from Charité, led by Geiger, have now conducted a study using rare tissue samples and advanced technology to show that the brain’s connectivity is not as previously assumed.
To conduct their study, the researchers analyzed brain tissue from 23 individuals who had undergone neurosurgery at Charité for drug-resistant epilepsy. During these surgeries, it was necessary to remove brain tissue in order to access the affected areas. This tissue was then used for the study. The researchers developed a clever method to listen in on neuronal communication within the brain.
The patients had agreed to allow the use of their diseased tissue for research purposes, including the structures beneath it.
In order to observe the signals passing between neighboring neurons in the outermost layer of the human neocortex, the team developed an improved version of the “multipatch” technique. This advancement allowed the researchers to monitor the communications between up to ten neurons simultaneously (for more information, refer to “About the method”). This enabled them to gather the necessary measurements to quickly map the network before the cells. “In humans, information tends to flow in one direction instead of in cycles,” explained Dr. Yangfan Peng, the first author of the publication. He worked on the study at the Institute for Neurophysiology and is now based at the Department of Neurology and Neuroscience. The researchers found that only a small fraction of the neurons engaged in reciprocal dialogue with each other. They analyzed the communication channels among nearly 1,170 neurons with about 7,200 possible connections.Research Center at Charité. The team utilized a computer simulation based on human network architecture to show that forward-directed signal flow has advantages in processing data.
The researchers tested the artificial neural network with a common machine learning task: identifying the correct numbers from audio recordings of spoken digits. The model that imitated human structures had a higher success rate in speech recognition compared to the mouse-modeled model. It was also more efficient, achieving the same performance with the equivalent of 380.The mouse model has around 800 neurons, while the human model only has about 150. Could AI be a model for the economy? “The network architecture in humans is more effective and efficient, as independent neurons can handle multiple tasks at the same time,” says Peng. “This means that the local network can hold more information. It’s still not clear if our findings in the outer layer of the temporal cortex apply to other cortical regions, or how well they can explain human cognitive abilities.” In the past, AI developers have drawn inspiration from biological models.When designing artificial neural networks, researchers have not only looked at the biological models but have also improved their algorithms independently. According to Geiger, many artificial neural networks already utilize forward-directed connectivity, which has been found to produce better results for certain tasks. The similarities between these network principles and the human brain are fascinating. Understanding the cost-efficient information processing in the human neocortex could offer valuable inspiration for enhancing AI networks.