Neural simulation pipeline: Enabling container-based simulations on-premise and in public clouds.

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Bioinformatics

Background:

• GENESIS is a powerful simulation engine, but struggles with the complexity of modern, large-scale neural network models.
• Setting up complex simulations involves managing software dependencies, parameters, and execution statistics, especially in High-Performance Computing (HPC) environments.
• Public cloud resources offer a scalable alternative to traditional on-premises clusters for computational neuroscience research.

Purpose of the Study:

• To present the Neural Simulation Pipeline (NSP) as a solution for managing and deploying complex neural simulations.
• To demonstrate the effectiveness of NSP using a pattern recognition task with the GENESIS engine and a custom visual system (RetNet).
• To evaluate NSP's performance and cost-effectiveness across different computing infrastructures, including on-premise and cloud environments.

Main Methods:

• Developed the Neural Simulation Pipeline (NSP) utilizing Infrastructure as Code (IaC) and containerization (Docker).
• Implemented a pattern recognition task using the GENESIS simulation engine and a biologically plausible spiking neuron model (RetNet).
• Executed 54 simulations across on-premise (HPI SOC Lab) and cloud (Amazon Web Services - AWS) infrastructures.

Main Results:

• NSP successfully facilitates large-scale neural simulations and deployment on diverse computing infrastructures.
• Containerized execution (Docker) demonstrated efficiency and portability.
• Cost analysis for AWS simulations was performed, highlighting cost-effectiveness.

Conclusions:

• The Neural Simulation Pipeline (NSP) significantly reduces the barriers to entry for conducting complex neural simulations.
• NSP enhances the practicality and cost-effectiveness of computational neuroscience research.
• The IaC and containerization approach enables flexible and scalable deployment of neural models.