Synaptic dynamics: linear model and adaptation algorithm.

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Systems Neuroscience

Background:

• Temporal processing in neural circuits is crucial for brain function.
• Existing models often lack detailed synaptic dynamics or biologically plausible learning rules.
• Understanding synaptic plasticity is key to neural computation.

Purpose of the Study:

• To develop a dynamic synapse neural network (DSNN) model for analyzing temporal processing.
• To propose a biologically plausible learning algorithm for synaptic adaptation in DSNNs.
• To evaluate the DSNN's performance in predicting neural activity and pattern recognition.

Main Methods:

• Introduced a linear approximate model for synaptic transmission dynamics, achieving >92.5% accuracy compared to non-linear models.
• Developed a supervised spike-in-spike-out learning rule for simultaneous pre- and post-synaptic component adjustment.
• Applied DSNNs to system identification, cortical neuron spiking prediction, and benchmark pattern recognition tasks.

Main Results:

• The linear synapse model simplifies analysis and training of spiking neural networks.
• The proposed learning algorithm demonstrated accuracy, repeatability, and scalability in system identification.
• DSNNs successfully predicted cortical neuron spiking patterns and outperformed other classifiers in pattern recognition with 96.7% accuracy.

Conclusions:

• The DSNN model provides an effective framework for understanding temporal dynamics in neural circuits.
• The proposed learning rule is biologically plausible and efficient for synaptic adaptation.
• DSNNs show significant potential for applications in neuroscience and AI, particularly in pattern recognition and neural data generation.