Embedding stochastic dynamics of the environment in spontaneous activity by prediction-based plasticity

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• The brain constructs internal models of the environment via sensory input, crucial for cognition.
• Spontaneous neural activity is increasingly recognized as reflecting these learned internal models.
• Existing computational models struggle to capture dynamic spontaneous activity matching sensory statistics.

Purpose of the Study:

• To propose biologically plausible synaptic plasticity rules for recurrent spiking neural networks.
• To enable the embedding of stochastic dynamics into spontaneous neural activity.
• To investigate how these rules facilitate learning internal models of the environment.

Main Methods:

• Developed a recurrent spiking neural network model.
• Introduced novel plasticity rules for excitatory and inhibitory synapses.
• Analyzed the statistical properties of spontaneous and stimulus-evoked activity.

Main Results:

• The proposed plasticity rules allow spontaneous activity to learn and reflect environmental statistical properties.
• Spontaneous cell assembly reactivation dynamics matched the model's evoked dynamics transition statistics.
• Simulations successfully replicated experimental findings on spontaneous activity in songbirds.

Conclusions:

• The developed plasticity rules offer a potential mechanism for learning internal environmental models.
• This approach bridges computational models of plasticity with observed spontaneous neural dynamics.
• The findings suggest a unified principle for internal model formation across different species.