Reinforcement Learning, Spike-Time-Dependent Plasticity, and the BCM Rule