Policy-based optimization: single-step policy gradient method seen as an evolution strategy

This research reports on the recent development of black-box optimization methods based single-step deep reinforcement learning and their conceptual similarity to evolution strategy (ES) techniques. It formally introduces policy-based (PBO), a policy-gradient-based algorithm that relies policy network describe density function its forthcoming evaluations, uses covariance estimation steer improvement process in right direction. The specifics PBO are detailed, connections evolutionary strategies discussed. Relevance is assessed by benchmarking against classical ES techniques analytic functions minimization problems, optimizing various parametric control laws intended for Lorenz attractor cartpole problem. Given scarce existing literature topic, this contribution definitely establishes as valid, versatile technique, opens way multiple future improvements building inherent flexibility neural networks approach.