Optimal percentage of inhibitory synapses in multi-task learning

Area of Science:

• Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Complex brains exhibit parallel task performance, characterized by excitatory and inhibitory synapses.
• Mammalian brains typically have 20-30% inhibitory synapses.
• Understanding the role of synapse ratios in learning is crucial for complex systems.

Purpose of the Study:

• To investigate parallel learning of Boolean rules in neuronal networks.
• To determine the optimal percentage of inhibitory synapses for multi-task learning.
• To explore the relationship between synapse ratios and network information processing.

Main Methods:

• Simulated neuronal networks performing multi-task learning of Boolean rules.
• Analysis of learning and forgetting dynamics.
• Quantification of network structure, including shortest path length.
• Varying the fraction of inhibitory synapses.

Main Results:

• Multi-task learning emerges from the alternation of learning and forgetting individual rules.
• A 30% fraction of inhibitory synapses optimizes overall network performance.
• This optimal fraction creates a complex backbone supporting efficient information transmission.
• 30% inhibitory synapses balance network excitability and resource-confining variability.

Conclusions:

• The optimal percentage of inhibitory synapses for complex learning is approximately 30%.
• This ratio supports efficient information processing and resource management in neuronal networks.
• Findings contribute to understanding the principles of complex brain function and artificial learning systems.