University course project on developing a spiking neural network implementation of a genetic algorithm.
Published here: https://link.springer.com/chapter/10.1007/978-3-030-76640-5_6
Authors: Siegfried Ludwig, Joeri Hartjes, Bram Pol, Gabriela Rivas
Center for Artificial Intelligence, Radboud University Nijmegen, The Netherlands
The use of evolution-inspired algorithms has been proven a viable solution for tackling problems of optimization. Considering the advantages in parallelism, memory collocation and energy efficiency of neuromorphic computing, it can be concluded that spiking neural networks could offer the possibility for the implementation of more efficient evolutionary algorithms. For this project, we designed and implemented scalable ensembles of spiking neurons to carry out the operations required for a genetic algorithm and implemented these in a simulator to successfully solve for an optimization problem. Two types of implementation have been explored that offer a complexity trade-off between computational space and time, with both designs having linear energy complexity.