Researchers from the University of Oxford have made a groundbreaking discovery about the learning efficiency of the human brain. In a study conducted by the MRC Brain Network Dynamics Unit and the Department of Computer Science at Oxford University, scientists have identified a new principle that explains how the brain adjusts connections between neurons during the learning process. This significant insight into the brain’s learning mechanism may revolutionize future research on brain networks and inspire the development of faster and more robust learning algorithms for artificial intelligence systems.
Traditionally, learning involves identifying the components in the information-processing pipeline that are responsible for errors in output. This concept is applied in artificial intelligence through a process called backpropagation, where a model’s parameters are adjusted to minimize output errors. Scientists have long speculated that the brain employs a similar learning principle.
However, the human brain demonstrates superior learning capabilities compared to current machine learning systems. Humans can learn new information with just one exposure, while artificial systems require repetitive training with the same information to achieve the same level of learning. Additionally, humans can acquire new knowledge while retaining existing knowledge, whereas artificial neural networks often struggle to integrate new information without interfering with or degrading existing knowledge.
These disparities led the researchers to investigate the fundamental principle underlying the brain’s learning process. By studying existing mathematical equations that describe the behavior of neurons and synaptic connections, the team analyzed and simulated various information-processing models. Their findings revealed a fundamentally different learning principle employed by the brain compared to artificial neural networks.
While artificial neural networks rely on external algorithms to modify synaptic connections and reduce errors, the researchers propose that the human brain first establishes an optimal balanced configuration of neuron activity before adjusting synaptic connections. This process, known as prospective configuration, is believed to be a highly efficient feature of human learning. By prioritizing the settling of neurons into an optimal configuration, the brain reduces interference and preserves existing knowledge, consequently accelerating the learning process.
The researchers published their findings in the prestigious journal Nature Neuroscience. In their study, they conducted computer simulations that demonstrated how models utilizing the prospective configuration principle outperformed artificial neural networks in tasks commonly encountered by animals and humans in natural environments.
To illustrate the concept, the researchers used the example of a bear fishing for salmon. The bear has learned that if it can see the river and simultaneously hear it and smell the salmon, catching a fish is likely. However, if the bear’s hearing is impaired, an artificial neural network model would also compromise the sense of smell, thus leading the bear to conclude that there are no salmon present and go hungry. In contrast, the animal brain maintains the knowledge of the salmon’s smell regardless of the lack of sound, allowing the bear to continue searching for fish.
The researchers have developed a mathematical theory supporting the notion that allowing neurons to settle into a prospective configuration reduces interference during the learning process. Through multiple learning experiments, they demonstrated that prospective configuration not only explains neural activity and behavior more effectively than artificial neural networks but also enhances learning speed and effectiveness.
While there is still a significant gap between theoretical models employing prospective configuration and our understanding of the brain’s anatomical networks, the researchers aim to bridge this gap through future research. Their goal is to explore the implementation of the prospective configuration algorithm in anatomically identified cortical networks and establish a comprehensive understanding of its workings.
Dr. Yuhang Song, the first author of the study, emphasized the need for specialized hardware or new types of computers that can rapidly and energy-efficiently implement prospective configuration, given the current technical limitations of simulating such processes on existing computers.
The groundbreaking discovery by the University of Oxford researchers not only sheds light on the brain’s remarkable learning abilities but also paves the way for advancements in artificial intelligence and machine learning. By unraveling the secrets of the human brain’s learning efficiency, scientists are unlocking new possibilities for creating more efficient and effective learning algorithms that can significantly enhance artificial intelligence systems in the future.
