Synaptic turnover promotes efficient learning in bio-realistic spiking neural networks

Area of Science:

• Neuroscience
• Artificial Intelligence
• Computational Neuroscience

Background:

• Machine learning (ML) systems require vast datasets and power for high performance.
• The human brain achieves complex cognitive tasks with remarkable energy efficiency.

Approach:

• Utilized a biologically constrained spiking neural network model.
• Investigated the role of synaptic turnover (structural plasticity) in neural computation and learning.

Key Points:

• Synaptic turnover significantly boosts network speed and performance on discrimination tasks.
• This mechanism enables accurate learning with reduced data requirements.
• Improvements are most pronounced under resource scarcity (e.g., reduced parameters, increased task difficulty).

Conclusions:

• Synaptic turnover is a key factor in the brain's efficient learning.
• Findings offer insights for developing more efficient and flexible ML algorithms.