Towards a learning-theoretic analysis of spike-timing dependent plasticity