Individual Neurons Can Learn by Predicting Future Activity
Humans have been trying to comprehend exactly how the brain works and how it obtains information for centuries. While neuroscientists now have a decent understanding of how different parts of the brain work and what their function is, numerous questions remain unanswered; therefore, a unified neuroscience theory is still missing.
In recent years, computer scientists have actually been attempting to create computational tools that artificially recreate the capabilities and processes of the human brain. New neuroscience theories shedding light on how the brain makes predictions can assist in substantially enhancing these tools to ensure that they reproduce neural functions in increasingly realistic ways.
The researchers at the Canadian Centre for Behavioural Neuroscience in Lethbridge, Canada, have recently executed a study investigating how individual neurons learn and predict the future. Their findings, released in Nature Machine Intelligence, suggest that the capacity of single neurons to anticipate their future activity can supply a new learning mechanism.
” Neuroscience is currently at the stage that biology was at prior to Darwin,” Artur Luczak, one of the scientists that carried out the study, informed TechXplore. “It has a myriad of comprehensive observations, yet no single theory explains the connections between them. Thus, the big pursuit in neuroscience is to find unifying principles to explain just how the brain works. Our work was aimed at adding to this pursuit.”
Utilizing mathematical equations, Luczak and his colleagues showed that the predictive capacities of single neurons could provide a new learning mechanism, which can eventually be replicated in machines. According to scientists, this learning process can have a metabolic origin, as neurons may need to reduce their very own synaptic activity while maximizing their influence on the local blood supply by enlisting other neurons.
Neurons and future activity prediction
” You learned that dark clouds foretell rainfall, as this aids you to stay dry and, therefore, conserve your heat,” Luczak explained. “Similarly, neurons may have the ability to learn that X quantity of input activity is generally followed by Y amount of activity. By changing synapses to decrease surprise– that is, the difference between real and expected activity– neurons can save energy by being active just as much as needed. We demonstrated that the predictive learning rule develops naturally due to maximizing metabolic energy by a neuron.”
In their paper, Luczak calls this learning mechanism the “lazy neuron principle.” The team is still uncertain about the specific mechanisms that might let a singular neuron to make forecasts, yet they think that they could be connected to calcium signaling (i.e., a procedure that involves the use of calcium ions to communicate and drive intercellular processes).
” Interestingly, our outcomes likewise indicate that spontaneous brain activity (e.g., during sleep) supplies ‘training data’ for neurons to learn to predict X from Y,” Luczak stated.
The recent research study carried out by this group of researchers might have several fascinating implications, both for the field of neuroscience and machine learning. In general, their discoveries suggest that a predictive mechanism underpinning the functioning of individual neurons can play an essential role in learning.
” In the future, this idea might additionally assist in creating more powerful artificial neural networks to fix tough real-life problems,” Luczak stated. “I believe that the predictive learning rule we unveiled is a vital step towards finding a unifying theory of the brain. However, more steps are needed to accomplish this, and we are delighted to proceed with this journey.
Read the original article on Tech Xplore.
Related “AI Shows that the Sahara has Millions of Trees”.
