Exact subthreshold integration with continuous spike times in discrete-time neural network simulations

Area of Science:

• Computational Neuroscience
• Neural Network Simulation
• Computational Efficiency

Background:

• Simulating large spiking neural networks efficiently often relies on time-grid constraints.
• Restricting spike times to a grid can lead to accuracy loss in neural dynamics.
• Interpolating spike times is explored to improve the accuracy of grid-based simulations.

Purpose of the Study:

• To develop and evaluate novel techniques for accurate and efficient simulation of large spiking neural networks.
• To combine exact integration schemes with off-grid spike event interpolation.
• To improve the computational efficiency and precision of neural network simulations.

Main Methods:

• Exact integration of subthreshold dynamics from one grid point to the next.
• Interpolation of spike times to identify off-grid events.
• Exploiting minimal synaptic propagation delay to remove central event queues.
• Developing a measure for simulation efficiency based on integration error.

Main Results:

• A novel method combining exact integration with off-grid spike interpolation was demonstrated.
• The need for central event queues was eliminated by leveraging synaptic propagation delays.
• For linear dynamics neuron models, local event queuing was also avoided, boosting single-neuron efficiency.
• The proposed techniques were shown to be more accurate and faster than standard methods across various input spike rates.

Conclusions:

• The novel techniques offer a more accurate and efficient approach to simulating large spiking neural networks.
• Combining grid-based exact integration with off-grid spike interpolation overcomes limitations of standard methods.
• These advancements are crucial for detailed computational neuroscience research and understanding complex neural dynamics.