Cross-entropy method

samples a batch of N values at each iteration, fits a Gaussian distribution to the best M < N of these samples, and then samples the next batch of N from that Gaussian