Maximizing Reliability with Bayesian Optimization
This work addresses reliability optimization in manufacturing, where failures are extremely rare, offering incremental improvements to Bayesian optimization methods.
The paper tackles the problem of maximizing design reliability under rare failure probabilities (10^-6 to 10^-8) by proposing two Bayesian optimization methods based on Thompson sampling and knowledge gradient, which outperform existing methods in both extreme and non-extreme regimes.
Bayesian optimization (BO) is a popular, sample-efficient technique for expensive, black-box optimization. One such problem arising in manufacturing is that of maximizing the reliability, or equivalently minimizing the probability of a failure, of a design which is subject to random perturbations - a problem that can involve extremely rare failures ($P_\mathrm{fail} = 10^{-6}-10^{-8}$). In this work, we propose two BO methods based on Thompson sampling and knowledge gradient, the latter approximating the one-step Bayes-optimal policy for minimizing the logarithm of the failure probability. Both methods incorporate importance sampling to target extremely small failure probabilities. Empirical results show the proposed methods outperform existing methods in both extreme and non-extreme regimes.