Göttingen, Germany – A groundbreaking advancement in artificial intelligence has emerged from the collaborative efforts of researchers at the Göttingen Campus Institute for Dynamics of Biological Networks (CIDBN) at the University of Göttingen and the Max Planck Institute for Dynamics and Self-Organization (MPI-DS). Scientists have developed a new type of artificial neuron, dubbed „infomorphic neurons,“ which learn independently and closely emulate the behavior of their biological counterparts in the human brain. The findings, published in the prestigious journal Proceedings of the National Academy of Sciences (PNAS), mark a significant step forward in bridging the gap between artificial and biological neural networks.
The human brain, a marvel of nature, and modern artificial neural networks share a common strength: their remarkable computational power. In both systems, neurons serve as the fundamental building blocks, processing information through interconnected networks. Traditional artificial neural networks, however, rely on a top-down approach to learning, requiring external coordination to adjust their behavior. This stands in stark contrast to biological neurons, which adapt and learn based solely on signals from neighboring neurons in their local environment. This localized, self-organized learning gives biological networks a significant edge in flexibility and energy efficiency—qualities that artificial systems have struggled to replicate.
Enter the infomorphic neuron: a revolutionary design that allows artificial neurons to learn autonomously, drawing inspiration directly from the brain’s pyramidal cells in the cerebral cortex. These artificial neurons no longer depend on external oversight to determine which inputs are relevant. Instead, they assess their immediate surroundings within the network and adjust their behavior accordingly. “We now directly understand what is happening inside the network and how the individual artificial neurons learn independently,” said Marcel Graetz, a researcher at CIDBN and a key contributor to the project.
The research team achieved this leap by redefining the learning process at the level of individual neurons. Using a novel information-theoretic measure, they programmed the infomorphic neurons to pursue clear, self-directed goals. These goals dictate whether a neuron should align its activity with its neighbors for redundancy, collaborate synergistically, or specialize in processing specific aspects of the input data. “By specializing in certain aspects of the input and coordinating with their neighbors, our infomorphic neurons learn how to contribute to the overall task of the network,” explained Valentin Neuhaus from MPI-DS.
This self-organized learning mirrors the adaptability seen in biological systems, where neurons dynamically respond to their local context. The result is an artificial neural network that not only performs tasks more efficiently but also offers insights into the brain’s own learning mechanisms. The researchers believe their work could pave the way for more advanced machine learning systems, capable of tackling complex problems with the same finesse and energy efficiency as the human brain.
Beyond its implications for artificial intelligence, the development of infomorphic neurons contributes to neuroscience by shedding light on how biological networks self-organize. As the team continues to refine this technology, their work promises to deepen our understanding of both artificial and natural intelligence, potentially revolutionizing fields from robotics to medical research. For now, this innovative step brings us closer to unlocking the secrets of the brain—and replicating its brilliance in silicon.
