Weight/neuron based learning rate

You can do that by defining different optimizers for each of the neurons. This link may help you.