Input–Output Uncertainty Comparisons for Discrete Optimization via Simulation
read more
Citations
Subsampling to Enhance Efficiency in Input Uncertainty Quantification
Classification and literature review on the integration of simulation and optimization in maritime logistics studies
Random perturbation and bagging to quantify input uncertainty
Stochastic approximation for simulation optimization under input uncertainty with streaming data
Fixed confidence ranking and selection under input uncertainty
References
Multivariate stochastic approximation using a simultaneous perturbation gradient approximation
Distributionally Robust Optimization Under Moment Uncertainty with Application to Data-Driven Problems
Robust Solutions of Optimization Problems Affected by Uncertain Probabilities
Robust Solutions of Optimization Problems Affected by Uncertain Probabilities
Related Papers (5)
Variable-fidelity model selection for stochastic simulation
Frequently Asked Questions (12)
Q2. What is the problem of interest in comparing k systems?
The problem of interest is to compare k systems, where the ith system’s performance measure isits simulation output mean, E[Yi(F c i )], under real-world input distribution F c i (c for correct), where Yi(·) is the stochastic output performance which depends on the chosen input distribution.
Q3. What is the condition for asymptotic normality of B/m(B?
Theorem 2 requires B = mγ for 0 < γ < 2, which is the condition for asymptotic normality of√ B/m(B̂i − Bi) in Proposition 3 in Section EC.5.
Q4. How is the performance measure estimated in a realistic DOvS setting?
In a realistic DOvS setting, each system’s performance measure is estimated via simulation replications, which introduces stochastic error.
Q5. How is the all-in procedure protected against such an error?
The all-in IOU-C procedure is protected against such an error by accounting for the estimation error in the gradients at the price of its conservatism.
Q6. How many values of (c) were sampled in the random search algorithm?
A total of L= 1,000 values of (θ̂−θc) were sampled in the random search algorithm (see Section EC.2) to approximate the optimal solutions of Pi`, i 6= `.
Q7. What is the average subset size of the plug-in procedure?
The average subset size of the plug-in procedure is 1.82, which is much smaller than that of all-in IOU-C, yet theestimated simultaneous coverage probability of the plug-in procedure is 0.874 (dashed line).
Q8. What is the average size of S0 for the conditional MCB procedure?
The average size of S0 is 1.03 for this procedure, which is the smallest among all three procedures since it ignores input uncertainty.
Q9. What is the simplest way to find the multidimensional quantile vectors?
Assumption 1(vii) states that given the plug-in distribution of CID effects and Vi(θ̂), the authors can find the exact multidimensional quantile vectors for −w(1)i` and −w(2)i` , respectively.
Q10. What is the definition of input model risk?
when wemake decisions based on the simulation outputs, the authors are subject to the risk of making suboptimaldecisions when the input models do not faithfully represent the real-world stochastic processes; thisis known as input model risk.
Q11. What is the coverage probability of the conditional procedure?
Figure 2 also shows that the simultaneous MCB coverage probabilityof the conditional procedure is 0.235 (dotted line), which is far lower than 0.9.
Q12. What is the probability of each unit arriving?
The actual number of units that arrive has a binomial distribution where the probability that each unit in the order arrives is 0.95.