Inferring plasticity rules from single-neuron spike trains using deep learning methods