Extend pooled noisy comparisons to other stochastic search methods

Determine whether pooled comparisons and the associated control of selection bias can be extended to localized random search, simulated annealing, genetic algorithms, and other stochastic search methods.

Background

The paper analyzes blind random search with noisy loss measurements using increasing averaging, a vanishing positive threshold, and reuse of accumulated measurements at the current estimate. Reusing measurements introduces selection bias because the current estimate and its pooled loss average depend on earlier acceptance decisions; the paper develops a uniform probabilistic bound to control this dependence and proves almost sure convergence under specified assumptions.

The authors leave unresolved whether the same pooled-comparison framework and selection-bias analysis can be adapted beyond blind random search, specifically to localized random search, simulated annealing, genetic algorithms, and other stochastic search methods. Such an extension would test the portability of the convergence strategy to algorithms with different candidate-generation and state-update mechanisms.