Dynamic synapsis in a neural network using a synaptic depression model.
This method shown in the code is applied only to a single hidden layer network. However, the plasticity effect will not be visible in the performance of the network unless it is applied to a multilayer network with more optimised architecture. The idea is to show how a simplified weight/gradient depression could work in a neural network.
The depression model is adapted from Paul Miller, An Introductory Course in Computational Neuroscience, MIT Press, 2019