Shallow Network Training With Dynamic Sample Weights Decay - a Potential Function Approximator for Reinforcement Learning | IEEE Conference Publication | IEEE Xplore