Adaptive Conditional Forest Sampling for Spectral Risk Optimisation under Decision-Dependent Uncertainty
Researchers have developed a new framework called Adaptive Conditional Forest Sampling (ACFS) to optimize spectral risk objectives, which combine expected cost and Conditional Value-at-Risk (CVaR). This method is particularly useful when the uncertainty distribution is decision-dependent, a scenario that complicates traditional modeling and simulation approaches. ACFS integrates several techniques, including Generalised Random Forests and guided exploration, to improve accuracy and reliability in estimating tail risks. Evaluations on benchmark datasets showed ACFS outperformed existing methods like GP-BO and CEM-SO in spectral risk reduction and run-to-run consistency. AI
IMPACT This research introduces a novel framework for improving risk management in complex, decision-dependent scenarios, potentially enhancing the reliability of AI-driven decision-making systems.