Dynamic synapse: a new concept of neural representation and computation

Area of Science:

• Neuroscience
• Computational Neuroscience
• Synaptic Plasticity

Background:

• Presynaptic mechanisms like facilitation and inhibition dynamically regulate neurotransmitter release probability.
• Synaptic strength in the hippocampus is not static but varies with the temporal pattern of action potentials.
• Recent findings indicate quantitative variations in these presynaptic dynamics across axon terminals of single hippocampal neurons.

Purpose of the Study:

• To explore the computational power of dynamic synapses with temporally varying release probabilities.
• To demonstrate how these dynamic synaptic properties contribute to neural information processing.
• To model and validate these dynamics using experimental data from hippocampal neurons.

Main Methods:

• Utilized a simple neural network model with synapses exhibiting dynamically determined release probability.
• Incorporated presynaptic mechanisms (facilitation, augmentation, inhibition) consistent with hippocampal synaptic transmission.
• Employed nonlinear systems analytic procedures to select model parameter values matching experimental data.

Main Results:

• The neural network successfully performed speech recognition on noisy raw waveforms from multiple speakers.
• Demonstrated that dynamic synapses possess significant representational and processing power for temporal spike patterns.
• Showcased an exponential increase in computational power when presynaptic dynamics vary across axon terminals.

Conclusions:

• Dynamic synapses, characterized by activity-dependent release probability, offer substantial computational advantages.
• These adaptable synaptic properties are crucial for processing complex temporal information and learning.
• The model's parameters align with experimental observations of hippocampal synaptic plasticity and dynamics.