From Wulfram Gerstner
Policy gradient algorithms give rise to three-factor rules. This video shows why.