Brian 2, an intuitive and efficient neural simulator.

Area of Science:

• Computational Neuroscience
• Artificial Neural Networks
• Scientific Simulation Software

Background:

• Simulating spiking neural networks (SNNs) is crucial for understanding brain function.
• Existing simulators often require low-level programming or lack expressiveness for complex experimental designs.
• Performance and reproducibility are key challenges in computational neuroscience model development.

Purpose of the Study:

• To introduce Brian 2, a simulator designed for simple and efficient SNN modeling.
• To address limitations of existing simulators regarding model complexity, expressiveness, and reproducibility.
• To demonstrate Brian 2's capability in handling novel dynamical equations, environmental interactions, and intricate experimental protocols.

Main Methods:

• Brian 2 employs runtime code generation to translate high-level descriptions into efficient low-level code.
• This method allows seamless integration of scientist-written code with the generated simulation code.
• The approach supports the definition of complex model dynamics, environmental interactions, and stimulation protocols.

Main Results:

• Brian 2 successfully simulates complex SNN models, including a plastic pyloric network and a closed-loop sensorimotor model.
• The simulator handles programmatic exploration of neuron models and real-time auditory input processing.
• Runtime code generation preserves high performance while enabling detailed and reproducible model descriptions.

Conclusions:

• Brian 2 offers a powerful solution for creating and simulating complex SNNs.
• The runtime code generation approach enhances ease of use, flexibility, and reproducibility in computational neuroscience.
• Brian 2 facilitates advanced research in neural dynamics, learning, and sensory processing.