AK-MCS-C2 : Active Kriging Monte Carlo Simulation method with conformal certification for failure probability estimation
Researchers have developed a new active-learning framework called AK-MCS-C2 that combines Active Kriging Monte Carlo simulation with conformal prediction for estimating failure probabilities. This method is particularly effective in small-sample settings and provides distribution-free guarantees on prediction errors, improving the accuracy and robustness of estimates for rare events. The framework utilizes an adaptive cross-conformal strategy and the J+GP conformal estimator with kriging surrogate models. AI
IMPACT This method offers improved uncertainty quantification for rare-event regimes, potentially benefiting AI systems that rely on accurate failure probability estimation.