Neurons violate rules during learning

The brain learns through multiple rules at the same time

Scientists have continued traditional wisdom about how we learn, revealing that individual neurons in the brain do not follow what had previously been believed to be a unified set of rules. This breakthrough finding suggests that different parts of the same neuron follow different rules while forming memory, potentially changing our understanding of learning disabilities and artificial intelligence.

Researchers at the University of California, San Diego use exquisite brain imaging techniques to visualize the learning of single synapses (connections between neurons) in real time. Their discovery, published April 17 in the journal Science, reveals a more complex and nuanced learning process than previously known.

“When people talk about synaptic plasticity, they are often considered unified,” said William “Jake” Wright, a postdoctoral scholar and lead author of the study. “Our study provides a clearer understanding of how synapses are modified during learning, with potentially important health effects, as many diseases in the brain involve some form of synaptic dysfunction.”

Learning is done as the brain’s neural network adapts, some connections strengthen, while others weaken, a process called synaptic plasticity. Although scientists have long studied the molecular mechanisms behind these changes, they have been working to understand why specific synapses are selected for modification, while others remain the same.

The researchers used cutting-edge two-photon imaging to observe mouse brains during learning tasks, allowing them to track changes at the level of a single synapse. The challenge they found was the basic assumptions about neural processing.

“This discovery fundamentally changes how we understand how the brain solves the credit allocation problem, where individual neurons perform different computations in parallel in different subcellular compartments,” explains senior author Takaki Komiyama, professor appointed in multiple departments at UC San Dieago.

“Credit allocation problem” refers to how a single synapse (only accesses local information) aggregates produce complex learning behaviors. This is similar to how a single ants perform a specific task without understanding the broader objectives of the colony.

This discovery can change the development of artificial intelligence. Current AI neural networks usually operate according to unified plasticity rules, but implementing multiple rules within a single unit may facilitate more complex systems that better mimic human learning.

These findings also provide promising avenues for the treatment of neurological and psychiatric disorders.

“This work is a potential basis for trying to understand how the brain usually works so that we can better understand what is wrong with these different diseases,” Wright noted.

Funded primarily by the National Institutes of Health, the research team is now investigating how neurons can simultaneously utilize different rules and the advantages of this multi-rule approach.