Learning Pareto-Efficient Decisions with Confidence
This work addresses decision support in safety-critical domains by providing statistically guaranteed trade-offs, though it is incremental as it builds on existing conformal prediction techniques.
The paper tackles the problem of multi-objective decision-making under uncertainty by extending Pareto efficiency to account for outcome uncertainty across contexts, enabling trade-off quantification for safety-critical applications. The result is a method for learning efficient decisions with statistical confidence, validated on synthetic and real data.
The paper considers the problem of multi-objective decision support when outcomes are uncertain. We extend the concept of Pareto-efficient decisions to take into account the uncertainty of decision outcomes across varying contexts. This enables quantifying trade-offs between decisions in terms of tail outcomes that are relevant in safety-critical applications. We propose a method for learning efficient decisions with statistical confidence, building on results from the conformal prediction literature. The method adapts to weak or nonexistent context covariate overlap and its statistical guarantees are evaluated using both synthetic and real data.