Neuromodulated Synaptic Plasticity on the SpiNNaker Neuromorphic System.

Area of Science:

• Computational Neuroscience
• Neuromorphic Engineering
• Machine Learning

Background:

• SpiNNaker is a low-power neuromorphic architecture for simulating large-scale spiking neural networks.
• Existing Spike-Timing-Dependent-Plasticity (STDP) rules on SpiNNaker support unsupervised learning but not environmentally-dependent behaviors.
• Neuromodulated STDP (three-factor learning rules) offer a biologically plausible mechanism for reinforcement learning.

Purpose of the Study:

• To implement a three-factor STDP learning rule, incorporating dopaminergic neuron feedback, on the SpiNNaker system.
• To demonstrate the rule's capability in simulating reward and punishment signals for synaptic plasticity.
• To investigate its application in solving the credit assignment problem in a Pavlovian conditioning task.

Main Methods:

• Development and implementation of a three-factor STDP model on the SpiNNaker neuromorphic architecture.
• Simulation of reward and punishment signals influencing synaptic plasticity at individual synapses.
• Large-scale network simulation of Pavlovian conditioning to address the credit assignment problem.

Main Results:

• Successful implementation of a three-factor STDP learning rule on SpiNNaker.
• Demonstration of reward/punishment signal delivery to synapses and network-level credit assignment.
• The three-factor STDP rule requires approximately twice the processing time of standard STDP but enables real-time simulation of up to 10,000 neurons.

Conclusions:

• The study presents the first implementation of a three-factor STDP learning rule on SpiNNaker.
• This enables biologically plausible reinforcement learning simulations on neuromorphic hardware.
• Opens new avenues for researching behavioral learning and memory mechanisms in large-scale neural networks.