[D] What do you think about representing each neuron and each synapse as a small neural network that will form a larger neural network?

For example, as input data, the synapse neural network will receive a forward propagation vector from the pre-neuron and a back propagation vector from the post-neuron. I think a model with a this approach can be taught to form short-term and long-term memory. Maybe we can use the output of the last gpt block to train this model.

back propagation data example form machinelearning network neural network neuron propagation small synapse think vector