Fast adaptation to rule switching using neuronal surprise.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Surprise is a key physiological response to unexpected events in humans and animals.
• Understanding the neuronal basis of surprise and its link to learning remains a challenge.

Purpose of the Study:

• To propose a self-supervised spiking neural network model that simulates the surprise signal.
• To investigate how surprise influences synaptic plasticity and learning.
• To address the stability-plasticity dilemma in neural networks.

Main Methods:

• Developed a self-supervised spiking neural network (SNN) model.
• Extracted a surprise signal from neural activity, specifically from excitation-inhibition imbalance.
• Implemented a three-factor learning rule modulated by the surprise signal to control synaptic plasticity.
• Utilized a modular network architecture to protect previously learned rules.

Main Results:

• The model successfully extracts a surprise signal correlating with rule switching.
• Synaptic plasticity is enhanced during moments of surprise, facilitating learning.
• The modular network design prevents overwriting of established rules, enhancing stability.
• The model demonstrates how surprise can be mechanistically linked to neuronal circuit dynamics.

Conclusions:

• The proposed SNN model provides a plausible mechanism for generating and utilizing surprise signals in the brain.
• This work offers insights into how the brain balances learning new information (plasticity) with retaining existing knowledge (stability).
• The model connects the subjective experience of surprise to concrete, testable predictions at the neural circuit level.