Neural networks reveal how experience shapes learning in both brains and machines
A new study has used a type of machine learning called a neural network to reveal how different kinds of training can change how learning happens—both in machines and in living brains.
A recent study has utilized neural networks to uncover how varied training methods can influence learning in both artificial intelligence and biological brains. Dr. Jack Bowler, a neurobiology postdoctoral fellow at the University of Utah Health, explained that neural networks can provide precise predictions regarding specific brain functions.
At a fundamental level, the analogy drawn by the researchers is that neural networks exhibit patterns of activity remarkably similar to those observed in real neurons within a crucial brain region associated with learning.
The study revealed that the method of training significantly impacts both the activity patterns and the capacity to solve tasks. For both neural networks and living brains, beginning with the acquisition of simpler skills enhances their ability to tackle complex tasks. Conversely, improper training could instill specific, predictable errors when confronted with more intricate challenges. The findings were published in Nature Neuroscience.
Researchers employed a specialized neural network task involving a go/no-go response based on the timing difference between two stimuli. This task, considered complex, parallels a mouse model experiment where mice learn to respond to specific olfactory patterns. When trained on the simpler go trials first, the neural networks demonstrated increased accuracy when transitioning to the full complex task.
In contrast, direct exposure to the complex task resulted in repetitive, foreseeable mistakes, such as premature responses during go trials.
The researchers further measured brain activity at the neuron level in mice during trials, focusing on a region linked to task learning. The observed patterns mirrored those found in the neural networks, suggesting that the models could serve as valuable tools for understanding human learning processes. This approach could expedite the generation of targeted hypotheses and minimize the number of animals required for research, ultimately benefiting the development of more effective training programs for humans and enhancing our comprehension of diseases impacting cognitive functions.
Written by urgent.news from Medical Xpress's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
