Approaching Quartic Convergence Rates for Quasi-Stochastic Approximation with Application to Gradient-Free Optimization

Neural Information Processing Systems 

Stochastic approximation is a foundation for many algorithms found in machine learning and optimization.