Reproducible Neural Network Simulations: Statistical Methods for Model Validation on the Level of Network Activity...

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Computational Science

Background:

• Neural network simulations are crucial for understanding brain dynamics.
• Lack of standardized validation hinders reproducibility in computational neuroscience.
• Existing formalisms do not cover practical validation workflows for neural simulations.

Purpose of the Study:

• To establish standardized statistical test metrics for validating neural network population dynamics.
• To introduce a practical workflow for iterative model validation.
• To address the need for generic, unbiased comparison of published neural models.

Main Methods:

• Development of standardized statistical test metrics for population dynamics.
• Introduction of a generic Python library for validation tests on neural activity data.
• Demonstration of an iterative model validation workflow using a spiking neural network on SpiNNaker hardware.

Main Results:

• A formal implementation of a verification and validation process for neural network simulations.
• Demonstration of iterative model validation for a spiking neural network.
• A Python library enabling quantitative validation of neural network activity data.

Conclusions:

• Standardized validation metrics are essential for reproducible computational neuroscience.
• The proposed workflow and library facilitate rigorous validation of neural network models.
• This work contributes to a consistent definition and implementation of verification and validation in neural simulations.