What is it about?
The bio-inspired spiking neural network (SNN) usually draws upon neurological evidence or theories to train its synaptic weights. Here, we used evolutionary algorithm to search for plasticity rules under a simple maze environment. Using various combinations of inputs, the discovered rules differ from familiar rules, yet were still able to solve the problem under the given constraints.
Featured Image
Photo by Jose Antonio Rodriguez Davia on Unsplash
Why is it important?
Neurological evidence has suggested many variations of the basic reward-modulated Spike-Time Dependent Plasticity (R-STDP) rules. Here, we test the idea that many more functions can be discovered through synthetic evolutionary optimisation. This allows the opportunity to create more customised learning rules for task-dependent applications.
Perspectives
Both the tasks (RL Environment) and the methods (SNN model, optimiser and fitness function) are still very simple, and could be expanded upon. Future research should focus on applying this methodology on more complex environments, more nuanced fitness functions, more diversity-driven algorithm and more realistic neuron models.
Napat Sahapat
Western Sydney University
Read the Original
This page is a summary of: EvoPlaSNN: Evolving Reward-modulated ANN-based Plasticity Rule for Spiking Neural Networks, July 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3795101.3805324.
You can read the full text:
Resources
Contributors
The following have contributed to this page
