A framework for plasticity implementation on the SpiNNaker neural architecture.

Area of Science:

• Computational Neuroscience
• Neuroscience
• Artificial Intelligence

Background:

• Understanding synaptic plasticity mechanisms is crucial for neural network simulations.
• Current hardware architectures struggle with large-scale simulations of plastic neural networks.
• Diverse biological plasticity phenomena necessitate flexible testing of computational models.

Purpose of the Study:

• To present a novel framework for investigating different synaptic plasticity approaches on the SpiNNaker platform.
• To leverage SpiNNaker's reconfigurable ARM processors for efficient plasticity simulations.
• To enable rapid testing of various learning rules before hardware commitment.

Main Methods:

• Developed a framework utilizing SpiNNaker's reconfigurable ARM processors for dedicated synaptic plasticity updates.
• Implemented and tested diverse spike- and rate-based learning rules, including STDP and BCM.
• Validated the framework by running real-time learning experiments on a 4-chip SpiNNaker board.

Main Results:

• Demonstrated the implementation of standard Spike-Timing Dependent Plasticity (STDP), voltage-dependent STDP, and BCM rule.
• Achieved efficient, modular, flexible, and scalable simulation of neural plasticity.
• Validated the framework's performance in classical learning experiments.

Conclusions:

• The proposed framework offers an efficient and flexible tool for exploring diverse neural learning models.
• It facilitates the rapid investigation of synaptic plasticity hypotheses on parallel and reconfigurable hardware.
• This work advances the simulation capabilities for large-scale plastic neural networks.